KR100567710B1 - Image coding apparatus, image coding method, image decoding method, image decoding apparatus, image data transmitting method and recording medium - Google Patents

Abstract

In order to produce a decoded picture that is substantially identical to the original picture by processing a small amount of data using a thinning circuit (subsampler), the level of compressed data obtained by extracting 9 pixels of the original picture into 1 pixel is 2 from the MSB. By extracting the bits, they are limited by the level limiting circuit. In the local decoder, the raw picture is predicted based on the correction data and the prediction values are output. Then, a prediction error of the prediction values for the raw image is calculated in the error calculator. The compressed data is corrected in the correction circuit so that the prediction error is reduced.

Description

IMAGE CODING APPARATUS, IMAGE CODING METHOD, IMAGE DECODING METHOD, IMAGE DECODING APPARATUS, IMAGE DATA TRANSMITTING METHOD AND RECORDING MEDIUM}

The present invention relates to an image coding apparatus, an image coding method, an image decoding method, an image decoding apparatus, a recording medium, and an image data transmission method, and in particular, it is possible to generate a high efficiency, high quality predictive image by performing less data processing operations. An image coding apparatus, an image coding method, an image decoding method, an image decoding apparatus, a recording medium, and an image data transmission method.

In the past, various image compression methods have been proposed. Among conventional image compression methods, there is an image compression method that compresses by thinning (subsampling) pixels of an image.

However, when the pixel data generated by simply thinning the image to compress the thinning image is directly coded, the total amount of data to be compressed becomes huge. Therefore, due to the huge processing demand, a high quality predictive image cannot be obtained effectively. Therefore, it is desirable to obtain a high quality, high efficiency predictive image using a small amount of data.

It is an object of the present invention to efficiently code image data.

It is a feature of the present invention to provide high picture quality using a small amount of data.

An advantage of the present invention is to reduce the amount of transmission or recording data required to provide high quality images.

According to one aspect of the invention, an apparatus for coding image data having a plurality of pixels comprises: means for generating compressed data by reducing the number of pixels of raw image data; Means for generating limited compressed data by limiting levels of pixels of the compressed data; Means for correcting the limited compressed data to produce corrected data; Means for predicting the raw image data according to the correction data to generate predictive data having a plurality of predicted pixels; Means for calculating a prediction error of the prediction data with respect to raw image data; Means for determining suitability of the correction data as coded data based on the prediction error; And means for outputting the correction data as coded data of the raw image data based on the determined suitability.

According to another feature of the invention, an apparatus for decoding coded data generated by coding image data comprises: means for receiving the coded data; And means for decoding the coded data to produce decoded picture data, wherein the coded data reduces the number of pixels of the raw picture data to produce compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; Repeating the correction step until the correction data becomes optimal correction data; And outputting the optimum correction data as coding data.

According to another feature of the invention, a method for coding image data having a plurality of pixels comprises the steps of: reducing the number of pixels of raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; And outputting the correction data as coded data of the raw image data based on the determined suitability.

According to another feature of the invention, a method for decoding coded data generated by coding image data comprises the steps of: receiving the coded data; And decoding the coded data to produce decoded picture data, wherein the coded data reduces the number of pixels of the raw picture data to produce compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; And outputting the correction data as coded data of the raw image data based on the determined suitability.

According to another feature of the invention, an apparatus for coding image data having a plurality of pixels comprises: a sub-sampling circuit for reducing the number of pixels of raw image data to generate compressed data; Limiting circuitry for generating limited compressed data by limiting levels of pixels of the compressed data; A correction circuit for correcting the limited compressed data to produce corrected data; A prediction circuit for predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; A calculator for calculating a prediction error of the prediction data with respect to the raw image data; Determination circuitry for determining suitability of the correction data as coded data based on the prediction error; And an output circuit for outputting the correction data as coded data of the raw image data based on the determined suitability.

According to another feature of the invention, an apparatus for decoding coded data generated by coding image data comprises: circuitry for receiving the coded data; And a decoder for decoding the coded data to produce decoded picture data, wherein the coded data reduces the number of pixels of the raw picture data to produce compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; And outputting the correction data as coding data of the raw image data based on the determined suitability.

According to one aspect of the invention, a method for transmitting image data having a plurality of pixels comprises the steps of: reducing the number of pixels of raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; Outputting the correction data as coding data of the raw image data based on the determined suitability; And transmitting the output correction data.

According to another feature of the invention, a recording medium has a coded digital signal recorded thereon, the recording medium comprising the steps of: reducing the number of pixels of raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Calculating a prediction error of the prediction data with respect to the raw image data; Determining suitability of the correction data as coded data based on the prediction error; Outputting the correction data as coding data of the raw image data based on the determined suitability; And recording the coded data on the recording medium.

According to another feature of the invention, a method for coding data having a plurality of samples comprises the steps of reducing the number of samples of raw data to produce compressed data; Limiting the level of samples of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate predictive data having a plurality of predicted samples; Calculating a prediction error of the prediction data with respect to the raw data; Determining the suitability of the correction data as coding data on the basis of the prediction error; And outputting the correction data as coding data of the raw data based on the determined suitability.

According to another feature of the invention, a method for decoding coded data generated by coding data comprises the steps of: receiving the coded data; And decoding the coded data to generate decoded data, the coded data reducing the number of samples of the raw data to produce compressed data; Limiting the level of samples of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate predictive data having a plurality of predicted samples; Calculating a prediction error of the prediction data with respect to the raw data; Determining the suitability of the correction data as coding data on the basis of the prediction error; And outputting the correction data as coding data of the raw data based on the determined suitability.

According to another feature of the invention, an apparatus for coding data having a plurality of fountains comprises: a sub-sampling circuit for reducing the number of samples of raw data to produce compressed data; Limiting circuitry for limiting the level of samples of the compressed data to produce limited compressed data; A correction circuit for correcting the limited compressed data to produce corrected data; A prediction circuit for predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction samples; A calculator for calculating a prediction error of the prediction data with respect to raw data; A decision circuit for determining suitability of the correction data as coding data on the basis of the prediction error; And an output circuit for outputting the correction data as coding data of the raw data based on the determined suitability.

According to another feature of the invention, an apparatus for decoding coded data generated by coding data comprises circuitry for receiving the coded data; And a decoder for decoding the coded data to produce decoded data, the coded data reducing the number of samples of the raw data to produce compressed data; Correcting the compressed data to generate correction data; Predicting the raw image data according to the correction data to generate predictive data having a plurality of predicted samples; Calculating a prediction error of the prediction data with respect to the raw data; Determining the suitability of the correction data as coding data on the basis of the prediction error; And outputting the correction data as coding data of the raw data based on the determined suitability.

According to an aspect of the present invention, an apparatus for coding image data having a plurality of pixels includes means for extracting a plurality of pixels of the image data to generate class information corresponding to characteristics of the extracted plurality of pixels. ; Means for storing mapping coefficients for each class; Means for reading mapping coefficients for each class; Means for reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And means for generating coded data by limiting the level of pixels of the correction data.

According to another feature of the invention, an apparatus for decoding coded data generated by coding image data comprises: means for receiving the coded data; And means for decoding the coded data to generate decoded picture data, wherein the coded data extracts a plurality of pixels of the picture data to generate class information corresponding to characteristics of the extracted plurality of pixels. Doing; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And generating the coded data by limiting the level of pixels of the correction data.

According to still another aspect of the present invention, a method for coding image data having a plurality of pixels includes extracting a plurality of pixels of the image data to generate class information corresponding to characteristics of the extracted plurality of pixels. ; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And generating coding data by limiting the level of pixels of the correction data.

According to another feature of the invention, a method for decoding coded data generated by coding image data comprises the steps of: receiving the coded data; And decoding the coded data to generate decoded picture data, wherein the coded data extracts a plurality of pixels of the picture data to generate class information corresponding to characteristics of the extracted plurality of pixels. step; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And generating the coded data by limiting the level of pixels of the correction data.

According to still another aspect of the present invention, a method for transmitting image data having a plurality of pixels includes extracting a plurality of pixels of the image data to generate class information corresponding to characteristics of the extracted plurality of pixels. ; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; Generating coded data by limiting levels of pixels of the correction data; And transmitting the coded data.

According to one aspect of the invention, a method for transmitting image data having a plurality of pixels comprises the steps of: extracting a plurality of pixels of the image data to generate class information corresponding to characteristics of the extracted plurality of pixels; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; Generating coded data by limiting levels of pixels of the correction data; And transmitting the coded data.

According to another feature of the present invention, a recording medium has a coded digital signal recorded thereon, and the recording medium extracts a plurality of pixels of the image data to obtain class information corresponding to characteristics of the extracted plurality of pixels. Generating; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; Generating coded data by limiting levels of pixels of the correction data; Transmitting the coding data; And recording the coded data on the recording medium.

According to still another aspect of the present invention, an apparatus for coding image data having a plurality of pixels is provided for extracting a plurality of pixels of the image data to generate class information corresponding to characteristics of the extracted plurality of pixels. Sub sampling circuit; A memory for storing mapping coefficients for each class; Circuitry for reading the mapping coefficients for each class; Circuitry for reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And a limiting circuit for generating the coded data by limiting the level of the pixels of the correction data.

According to another feature of the invention, an apparatus for decoding coded data generated by coding image data comprises: circuitry for receiving the coded data; And a decoder for decoding the coded data to generate decoded picture data, wherein the coded data extracts a plurality of pixels of the picture data to generate class information corresponding to characteristics of the extracted plurality of pixels. Doing; Storing mapping coefficients for each class; Reading the mapping coefficients for each class; Reading mapping coefficients corresponding to the class information and generating correction data using the image data and the read mapping coefficients; And generating the coded data by limiting the level of pixels of the correction data.

According to another feature of the invention, a method of transmitting coded picture data comprises the steps of: receiving the coded picture data; And transmitting the coded picture data, wherein the coded picture data reduces the number of samples of the raw data to produce compressed data; Limiting the level of samples of the compressed data to generate limited compressed data; Correcting the limited compressed data to produce corrected data; Predicting the raw image data according to the correction data to generate predictive data having a plurality of predicted samples; Calculating a prediction error of the prediction data with respect to the raw data; Determining the suitability of the correction data as coding data on the basis of the prediction error; And outputting the correction data as coding data of the raw data based on the determined suitability.

The above and other objects, features and advantages of the present invention will become apparent upon reference to the following description and the accompanying drawings.

BRIEF DESCRIPTION OF DRAWINGS To clarify the corresponding relationship between the respective means of various embodiments described herein, some features of the present invention will first be described briefly with reference to the drawings. The term "unit" herein is to be interpreted in a broad sense, including hard wired circuits, mainframes with appropriate software, programmed microprocessors or micro-controllers, or combinations thereof.

1 illustrates a configuration of an embodiment of an image processing apparatus to which the present invention is applied. The digitized image data is supplied to the transmission device 1. The transmission device 1 compresses / codes thinned image data to obtain coded image data by thinning the input image data (i.e., reducing the number of pixels by subsampling, for example), Record this coded data on recording media 2, such as optical discs, magneto-optical discs, magnetic tapes, and other recording media, or otherwise code this data over terrestrial, satellite, telephone, CATV networks, or other Send through the path.

In the reception device 4, coded data recorded on the recording medium 2 is reproduced, or coded data transmitted through the transmission path 3 is received, and the coded data is expanded and decoded, A decoded picture is formed and fed to a display (not shown) and displayed on the display.

The image processing apparatus described above can be used, for example, in an image recording / reproducing apparatus such as an optical disk apparatus, a magneto-optical disk apparatus, a magnetic tape apparatus or the like. Moreover, such an image processing apparatus can also be used for image transmission apparatuses, such as a television telephone apparatus, a television broadcasting system, and a CATV system. Further, since the data amount of the coded data output from the transmission device 1 is small, the image processing device of FIG. 1 can be used for a device having a low transmission speed, for example, a mobile phone and other portable terminals.

2 shows an example of the configuration of the transmission device 1 described above.

An I / F (interface) 11 is provided to perform reception processing of image data supplied from the outside and processing of coded data for the transmission / recording device 16. Read Only Memory (ROM) 12 stores programs for Initial Program Loading (IPL) (initial program loading) and other programs. Random Access Memory (RAM) 13 is a system program (i.e., an operating system), an application, and data necessary for the operation of the central processing unit (CPU) 14 stored in the external storage device 15. It is used to store it. The CPU 14 expands the system program and the application program from the external storage device 15 to the RAM 13 according to the IPL program stored in the ROM 12, and also executes the application program under the control of this system program. The decoding processing operation (described later) is executed on the image data supplied from the F 11. The external storage device 15 is composed of, for example, a magnetic disk device. As described above, the external storage device 15 stores application programs and system programs executed by the CPU 11, and also stores data necessary for the operation of the CPU 14. Save it. The transmission / recording device 16 records the coded data supplied from the I / F 11 on the recording medium 2 or transmits the coded data via the transmission path 3.

Note that the above-described I / F 11, ROM 12, RAM 13, CPU 14, and external storage device 15 are connected to each other via a bus 17.

In the transmission device 1 configured as described above, when image data is supplied to the I / F 11, such image data is supplied to the CPU 14. The CPU 14 codes the image data and then supplies the coded data to the I / F 11. As soon as the I / F 11 receives the coded data, it supplies it to the transmission / recording device 16. In the transmission / recording device 16, the coded data derived from the I / F 11 is recorded on the recording medium 2 or transmitted via the transmission path 3.

First embodiment

Now, the configuration of the first embodiment of the present invention will be described with reference to FIGS. 3 to 6, 8, and 10 to 15. FIG. 3 is a functional block diagram showing an internal configuration of the transmission device 1 of FIG. 2 except for the transmission / recording device 16. As shown in FIG.

The picture data (raw picture data) to be coded is supplied to the compression unit 21, the local decoder 22, and the error calculator 23. The compression unit 21 compresses the image data by simply thinning (subsampling) the pixels of the image data, and restricts the level of the extracted pixel data, thereby controlling such compressed data (i.e., counting three pixels). Linearized and level limited image data). The correction data obtained by correcting the compressed data of the compression unit 21 is supplied to the local decoder 22 and the determination unit 24.

The local decoder 22 predicts the original image based on the correction data supplied from the compression unit 21, and then supplies the predicted value to the error calculator 23. As will be described later, the local decoder 22 performs an appropriate processing operation to obtain the prediction coefficients used to calculate the prediction value by linear coupling with the correction data. Based on this prediction coefficient, the local decoder 22 obtains the prediction values, which in turn are supplied to the error calculator 23. The resulting prediction coefficient is supplied to the determination unit 24.

The error calculator 23 calculates the prediction error of the predicted value derived from the local decoder 22 with respect to the raw image data (raw image) flowing into the error calculator 23.

Based on the error information derived from the error calculator 23, the determination unit 24 determines whether or not the correction data output from the compression unit 21 exactly matches the coded result of the original image. Then, if the determination unit 24 determines that the correction data output from the compression unit 21 does not match the coded result of the original image, the determination unit 24 controls the compression unit 21 to control the compressed data. Is corrected again so that new correction data is output from this compression unit 21. In addition, when the determination unit 24 determines that the correction data output by the compression unit 21 exactly matches the coded result of the raw image data, the correction data supplied from the compression unit 21 is optimized for compression data ( Hereinafter, if necessary, " optimal compressed data ") is supplied to the multiplexer 25, and the prediction coefficient supplied from the local decoder 22 is supplied to the multiplexer 25.

The multiplexer 25 multiplexes the optimal compressed data derived from the determination unit 24 with the prediction coefficients, and then supplies the multiplexed result to the transmission / recording device 16 as coded data (Fig. 2).

Next, the operation of the transmission device of FIG. 3 will be described again with reference to the flowchart of FIG. 4. When image data (raw image data) is supplied to the compression unit 21, the compression unit 21 first compresses this image data, while thinning the image data in step S1 (i.e., a plurality of pixels). Extract some pixels from the In step S2, the level of extracted pixel data is limited by extracting two bits of pixel data on the MSB side of such pixel data. This pixel data is then output to the local decoder 22 and the determination unit 24 without any correction initially. In the local decoder 22, in step S3, the correction data derived from the compression unit 21 is locally decoded. This correction data is the same as the compressed data, so that, as described above, the image data is simply thinned and initially level controlled.

That is, in step S3, an appropriate processing operation is performed to obtain prediction coefficients used to calculate the predicted value of the raw image data by the linear coupling operation with the correction data from the compression unit 21. Next, based on these prediction coefficients, a prediction value is calculated. The predicted value calculated at the local decoder 22 is supplied to the error calculator 23, and the predictive coefficient is supplied to the determination unit 24.

In this case, the picture composed of the predicted values output from the local decoder 22 is the same as the decoded picture obtained on the receiving device 4 side (Fig. 1).

If the predicted value of the original image data is received from the local decoder 22 by the error calculator 23, in step S4, the error calculator 23 derives the predicted value derived from the local decoder 22 with respect to the original image data. The prediction error of is calculated, and this prediction error is supplied to the determination unit 24 as error information. When the error information is received from the error calculator 23, the determination unit 24 determines whether or not the correction data output from the compression unit 21 in step S5 exactly matches the coded result of the original image data. Judge for.

That is, in step S5, for example, a determination is made as to whether or not the error information is less than the predetermined threshold value "?". In step S5, if it is determined that the error information is not less than the predetermined threshold value "ε", it is recognized that the correction data output from the compression unit 21 does not match the coded data of the original image data. Subsequently, the processing operation proceeds to step S6 in which the determination unit 24 controls the compression unit 21 to correct the compressed data. The compression unit 21 changes the correction amount (that is, the correction value "Δ" will be described later) under the control of the determination unit 24 to correct the data. Finally, the corrected data is output to the local decoder 22 and the determination unit 24. Then, the processing operation returns to the previous step S3, and a similar processing operation is repeatedly executed.

On the other hand, if it is determined in step S5 that the error information is less than the predetermined threshold value "ε", it is recognized that the correction data output from the compression unit 21 exactly matches the coded result of the raw image data. The determination unit 24 outputs the correction data when the error information is smaller than the predetermined threshold value "ε" to the multiplexer 25 together with the prediction coefficient as the optimum correction data. In step S7, the multiplexer 25 multiplexes the optimal compressed data and prediction coefficients derived from the determination device 24, and outputs this multiplexed result as coded data. Thereafter, the processing operation ends.

As described above, in the state where the error information becomes less than the preselected threshold value "ε", since the correction data generated by correcting the compressed data is used as a coded result of the original image data, the receiving device 4 Based on this corrected data on the side, an image substantially identical to the original image can be obtained.

Next, FIG. 5 illustrates the configuration of the compression unit 21 of FIG. 3.

The image data (raw image data) to be coded is input into the thinning unit 30. The thinning unit 30 can reduce the input image data to 1 / N. As a result, compressed data in which image data is compressed at 1 / N is output from the thinning unit 30. Such compressed data is introduced into the level limiting device 31 to limit the level of such compressed data, for example, by extracting only 2-bit data of each pixel data on the MSB side of the compressed data, and such pixel data is represented by 8 bits. . The level limited compressed data is then supplied to the correction unit 32.

In response to the control signal sent from the determination unit 24, the correction unit 32 supplies an address to the correction value ROM 33, so that the correction value " Δ " is read from this ROM. Subsequently, the correction unit 32 generates the correction data by, for example, adding the correction value Δ read from the correction value ROM 33 to the compressed data from the level limiting unit 31, and then generates the correction data generated. Is supplied to the local decoder 22 and the determination unit 24. The correction value ROM 33 stores a combination of various correction values Δ (for example, a combination of correction values for correcting compressed data for one frame) and stores compressed data output from the level limiting unit 31. Correct it. The combination of correction values Δ corresponding to the address supplied from the correction unit 32 is read out from this correction value ROM 33 and supplied to the correction unit 32.

For example, when image data for one frame is supplied to the correction unit, in step S11, the thinning unit 30 thins this image data in 1 / N, and the resulting compressed data is transferred to the level limiting device 31. Output

In this case, as shown in Fig. 7, the thinning unit 30 is designed to thin the image data, for example to 1/9. That is, the thinning unit 30 extracts only one pixel value (a portion indicated by the symbol " " in Fig. 7) around the center pixel in each unit (here, 9 of 3 x 3 (horizontal direction x vertical direction) The pixel is used in units of 1), and other portions (parts indicated by the symbol " " Here, it should be noted that the thinning unit 30 performs the above-described processing operation in units of one frame, for example. As a result, the data generated by thinning the image data for one frame in 1/9 is supplied from the thinning unit 30 to the level limiting device 31 as compressed data. Further, the thinning processing operation performed in the thinning unit 30 may be performed such that an image of one frame is subdivided into several blocks, and the thinning processing operation may be performed in units of one block. Be careful. Upon receiving the supply of thinned pixel data 1/9 from the thinning unit 30, in step S12, the level limiting device 31 thins out by extracting 2-bit data from the pixel data indicated by 8 bits from the MSB side. Limit the level of pixel data. That is, as a result, since each pixel data is represented by 2 bits, subsequent processing operations can be simplified.

Upon receiving the compressed data from the level limiting device 31, the correction device 32 determines whether or not the control signal transmitted from the determination device 24 (FIG. 3) has been received. If it is determined in step S13 that the control signal has not yet been received, the process proceeds to step S16 in which the correction device 32 uses the compressed data derived from the level limiting device 31 as correction data and the local decoder 22. And directly to the judging device 24. The processing operation then returns to step S13.

That is, as described above, the determination device 24 controls the compression device 21 (correction device 32) in accordance with the error information. Immediately after receiving the compressed data in the level limiting device 31, error information is not yet obtained (since no error information is output from the error calculator 23), so that no control signal is output from the judging device 24. As a result, immediately after the compressed data is output from the level limiting device 31, the correcting device 32 cannot correct the compressed data (by adding 0 to correct it), and converts the compressed data as the corrected data to the local decoder ( 22) and directly to the determination device 24.

On the other hand, when it is determined in step S13 that the control signal is received from the determination device 24, the correction device 32 outputs the address determined in accordance with this control signal to the correction value ROM in step S14. As a result, in this step S14, the combination (set) of the correction values Δ stored at this address and used to correct the compressed data for one frame is read out from the correction value ROM 33, and then the correction device ( 32). Upon receiving a combination of correction values? Read from correction value ROM 33, correction device 32 compresses one frame at step S15 by adding the corresponding correction value? To respective compressed data for one frame. Compute the data generated by calibrating the data. Thereafter, the process proceeds to step S16, where correction data is output from the correction device 32 to the local decoder 22 and the determination device 24, and then the flow returns to step S13. If it is determined in step S13 that the control signal having the end of the operating state has been received from the judging device 24, the process returns to step S11, where the thinning unit 30 returns the image data in step S11 to 1 / N. Thinning, and then processing subsequent compressed data in the same manner.

As described above, the compression device 21 repeatedly outputs the corrected data obtained by correcting the compressed data to various values under the control of the determination device 24.

Note that when the coding operation for the image of one frame is completed, the determination device 24 supplies the compression device 21 with a control signal indicating the completion of this coding operation. Upon receiving such a control signal, the compression device 21 executes the processing operation specified in the flowchart of Fig. 6 relating to the image of the next frame.

In addition, in the above-described case, the compressed data is generated by the thinning unit 30 by extracting only pixel data (pixel values) near the center pixel of 3x3 pixels. As an alternative, for example, an average value of 3x3 pixels is calculated, and thus compressed data can be generated by using this average value as the pixel value of the center pixel in the 3x3 pixel.

Next, FIG. 8 shows the configuration of the local decoder 22 shown in FIG.

The corrected data derived from the compression device 21 is supplied to the block generation unit 41 for class classification and the block generation unit 42 for predictive value calculation. The class classification block generation unit 41 makes a class classification block centered on the displayed correction data into units used to classify the corrected data into predetermined classes according to its properties.

That is, for example, the corrected data (compressed data) (or pixels) (parts indicated by the symbol "●") located at the i-th position from the upper position and the j-th position from the left position in FIG. 7 is represented by "X _ij ". Then, the class classification block generating unit 41 has a total of five pixels, that is, one pixel "X _ij " of interest and adjacent to this pixel of interest in the upper, left, right, and lower positions. It constitutes a class classification block arranged in four pixels, that is, X _{(i-1) j} , X _{i (j-1)} , X _{i (j + 1)} , and X _{(i-1) j} . This class classification block is supplied to a class classification adaptive processing circuit 43.

Note that in this case, the class classification block is composed of cross-shaped blocks, but it is not necessary to be such a cross. As an alternative, for example, rectangular, square, or any other shape may be used. In addition, the total number of pixels for constructing the class classification block is not limited to five pixels.

The predictive value calculation block generating unit processes the corrected data to make the predictive value calculation block centered on the displayed corrected data as a unit used for calculating the predicted value of the original image data. That is, while positioning the corrected data X _ij (the portion indicated by the symbol ".&Quot; in FIG. 7), the pixel values of 9 (= 3 × 3) pixels in the original image data in FIG. 7 move from the leftmost direction to the rightmost direction. And Y _ij (1), Y _ij (2), Y _ij (3), Y _ij (4), Y _ij (5), Y _ij (6), Y _ij (7), Y _ij (8), assuming that represented by Y _ij (9), the predicted value calculation block generating unit 42 calculates a predicted value of the squares to calculate the predicted values of the pixels Y _ij (1) through the pixel Y _ij (9) for It will construct a block. This predictive value calculation block is for example the following 25 (= 5 x 5) pixels, i.e., X _{(i-2) (j-2)} , X _{(i-2) (j-1)} , X _{(i-2) j} , X _{(i-2) (j + 1)} , X _{(i-2) (j + 2)} , X _{(i-1) (j-2)} , X _{(i-1) (j-1)} , X _{(i-1) j} , X _{(i-1) (j + 1)} , X _{(i-1) (j + 2)} , X _{i (j-2)} , X _{i (j-1)} , X _ij , X _{i (j + 1)} , X _{i (j + 2)} , X _{(i + 1) (j-2)} , X _{(i + 1) (j-1)} , X _{(i + 1) j} , X _{( i + 1) (j + 1)} , X _{(i + 1) (j + 2)} , X _{(i + 2) (j-2)} , X _{(i + 2) (j-1)} , X _{(i + 2) j} , X _{(i + 2) (j + 1)} , X _{(i + 2) (j + 2)} . Here, the pixel X _ij is set to the center.

Specifically, for example, in order to calculate the predicted values of the pixels Y33 (1) to Y33 (9) in the original image data enclosed by the rectangles in FIG. 7, the predicted value calculation block may include the following pixels, that is, X ₁₁ , X ₁₂ , X ₁₃ , X ₁₄ , X ₁₅ , X ₂₁ , X ₂₂ , X ₂₃ , X ₂₄ , X ₂₅ , X ₃₁ , X ₃₂ , X ₃₃ , X ₃₄ , X ₃₅ , X ₄₁ , X ₄₂ , X ₄₃ , X ₄₄ , It consists of X ₄₅ , X ₅₁ , X ₅₂ , X ₅₃ , X ₅₄ , X ₅₅ .

The predictive value calculation block obtained by the predictive value calculation block generation unit 42 is supplied to the class classification adaptive processing circuit 43.

Similarly to the class classification block, the number and shape of the pixels for the predictive value calculation block are not limited to the above-described example. Note that it is preferable that the total number of pixels for constructing the predictive value calculation block is larger than the total number of pixels for constructing the class classification block.

In addition, when the above-described block generation processing is performed (similar to the processing operation other than block generation), there is no corresponding pixel near the image frame. In this case, for example, assuming that the same pixel as the pixel constituting the image frame is located outside this image frame, the processing operation can be performed.

The class classification adaptation processing circuit 43 is arranged by the class classification unit 45 and the adaptation processing unit 46, and is designed to perform the class classification adaptation processing operation.

The class classification adaptive processing operation means that an input signal is classified into several classes according to its characteristics, and then an adaptive processing operation suitable for each of these classes is performed on the input signal of each of these classes. Briefly described, this class classification adaptive processing operation is subdivided into class classification processing operation and adaptive processing operation.

The class classification processing operation and the adaptive processing operation will now be briefly described.

First, the class classification processing operation will be described.

Here, for example, as shown in Fig. 9A, a block of 2x2 pixels (class classification block) is composed of a target pixel and three pixels positioned adjacent thereto, each of which has one bit (level "0" and level). It is assumed that the expression "1" has any one. In this case, as shown in FIG. 9B, blocks of 4 (2 × 2) pixels may be classified as 16 (= (2 ₁ ) ₄ ) according to the level distribution of each pixel. Such pattern classification corresponds to the class classification processing performed in the class classification unit 45.

Note that the class classification processing operation can be performed in consideration of the activity (image complexity, intensity of change) of the picture (picture in the block).

Typically, in this example, for example, eight or more bits are allocated to each pixel. In addition, in the present embodiment, as described above, the class classification block is arranged in 5 pixels. As a result, when performing a class classification processing operation on such a class classification block composed of original (8-bit) pixels, this block is classified into such a huge number (2 ⁸ ) ⁵ classes.

Therefore, according to the present embodiment, as described above, since the number of bits of the pixel is reduced from 8 bits to 2 bits in the level limiting device 31, the number of classes is deleted (1024 (= (2 ² ) ⁵ ).

Subsequently, the adaptive processing operation will be described.

For example, it is assumed that the predicted value E [y] of the pixel value " y " of the image is calculated with a linear primary coupling. This linear primary coupling model uses the pixel values X ₁ , X ₂ ,... (Hereinafter referred to as "learning data" if necessary) and predetermined prediction coefficients W ₁ , W ₂ ,. It is defined by the linear combination of. In this case, the predicted value E [y] can be expressed by the following equation.

E [y] = W ₁ X ₁ + W ₂ X ₂ +... + W ₂₅ X ₂₅ + WO

Here, the symbol " WO " is a compensation coefficient added to represent E [y] by 8 bits, because the original 8 bit data " X _i " is converted into 2 bit data.

Therefore, in order to generalize this equation, if the matrix W of the prediction coefficient W set, the matrix X of the training data set, and the matrix X of the prediction value E [y] set are defined by Equation 1, the following observation is satisfied. Can be.

Next, the predicted value E [y] approximating the pixel value y of the original image data is considered to be obtained by applying the least square method to this observation. In this case, the matrix Y of the original image data (hereinafter referred to as "teacher data", if necessary), and the other of the rest of the set of prediction values E [y] for the pixel values y of the image If the matrix E is defined by equation (2), the following remainder can be satisfied from equation (2).

을 최소화함으로써 계산될 수 있다.In this case, the prediction coefficient Wi used to calculate the prediction value E [y] approximating the pixel value y of the original image data is the squared error.

Can be calculated by minimizing

As a result, the square error differentiated by the prediction coefficient Wi becomes zero. That is, the prediction coefficient Wi, which can satisfy the following equation, is an optimal value used to calculate the prediction value E [y] that is approximated to the pixel value "y" of the original image data.

Therefore, first, the following formula can be satisfied because Equation 3 is differentiated by the prediction coefficient Wi.

According to Equations 4 and 5, Equation 6 can be obtained.

Furthermore, considering the relationship between the learning data "x", the prediction coefficient "w", the teacher data "y", and the remaining "e" contained in the remaining equations defined in equation (3), You can get it.

A regular expression defined in Equation 7 can be established, and the number of regular expressions is equal to the number of prediction coefficients w to be calculated. As a result, the optimal prediction coefficient w can be obtained by solving equation (7). Note that, for solving equation (7), for example, a sweep-out method (Gaussian-Jordan erase method) may be applied.

As described above, the adaptive processing operation is executed such that the optimum prediction coefficient w and the prediction value E [y] approximating the pixel value y of the original image data are calculated according to the equation (1) by using this prediction coefficient w. This adaptation processing operation is performed in the adaptation processing unit 46.

Note that this adaptive processing operation is different from the interpolation processing operation. That is, in this adaptive processing operation, components contained in the original image data but not included in the thinned image data are reproduced. That is, the adaptive processing operation is the same as the interpolation processing operation employing the so-called "interpolation filter" considering only the equation (1). However, since the prediction coefficient w corresponding to the tap coefficient of this interpolation filter is calculated using the teacher data "y", i.e., the learning system, the components contained in the original image data can be reproduced. Considering this technical point of view, it can be said that the adaptive processing operation is a processing operation having an image forming effect.

The operation of the local decoder 22 of FIG. 8 will now be further described with reference to the flowchart of FIG. 10.

First, in the local decoder 22, the corrected data derived from the compression device 21 in step S21 is processed to make a block. That is, in the class generating block generation unit, the correction data is subdivided into class classification blocks of five pixels around the displayed corrected data, which will be supplied to the class classification adaptive processing circuit 43. In the predictive value calculating block generating unit 42, the corrected data is subdivided into predicted value calculating blocks of 5x5 pixels around the displayed corrected data, which will be supplied to the class classification adaptive processing circuit 43.

As described above, in addition to the class classification block and the predictive value calculation block, raw image data is supplied to the class classification adaptive processing circuit 43. The class classification block is supplied to the class classification unit 45, and the predictive value calculation block and the raw image data are supplied to the adaptive processing unit 46.

In step S22, the class classification unit 45 performs a class classification processing operation on the class classification block. That is, the class classification unit 45 detects the distribution state of each of the pixel levels in the block, and then determines which class this class classification block belongs to. This class determination result is supplied to the adaptive processing unit 46 as class information.

In this embodiment, since the class classification processing operation is performed on a class classification block composed of five pixels represented by two bits, each of the class classification blocks is classified into one of 1024 classes (= (2 ² ) ⁵ ). You will know.

Thereafter, the processing proceeds to step S23 to perform an adaptive processing operation for each class based on the class information supplied from the class classification unit 45 to the adaptive processing unit 46. As a result, both the prediction value and the prediction coefficient for each class of the original picture 1 frame are calculated.

That is, according to this embodiment, for example, when paying attention to any correction data, 25x9 prediction coefficients for each class are calculated using the correction data and the raw image data for one frame, and the 25x 9 prediction coefficients and a predictive value calculation block composed of 5x5 pixels centered on the corrected data noted in response to class information are executed to execute the adaptive processing operation, so that a total of 5 pixels, i.e., the corrected corrected data Prediction values for one corresponding pixel and four pixels adjacent to the pixel are calculated.

Specifically, for example, class information C is output from the class classification unit 45, and this class information C includes five correction data X ₂₃ , X ₃₂ , centered on correction data X ₃₃ shown in FIG. 7. Assume that it relates to a class classification block composed of X ₃₃ , X ₃₄ , and X ₄₃ . It is also assumed that the predicted value calculating block is output from the predicted value calculating block generating unit 42 as the predicted value calculating block corresponding to the class classification block. This predictive calculation block is X ₁₁ , X ₁₂ , X ₁₃ , X ₁₄ , X ₁₅ , X ₂₁ , X ₂₂ , X ₂₃ , X ₂₄ , X ₂₅ , X ₃₁ , X ₃₂ , X centered on correction data X ₃₃ . ₃₃ , X ₃₄ , X ₃₅ , X ₄₁ , X ₄₂ , X ₄₃ , X ₄₄ , X ₄₅ , X ₅₁ , X ₅₂ , X ₅₃ , X ₅₄ , X ₅₅ . First, the correction data constituting this predictive value calculation block is set as the training data, and pixel values Y ₃₃ (1) to Y ₃₃ (9) of 3x3 pixels centering on the correction data X ₃₃ in the initial image data (Fig. By setting the area enclosed by the rectangle at 7) as teacher data, the normal equation shown in equation (7) can be established.

Also, for example, for one frame, a normal equation is similarly established for other prediction value calculation blocks corresponding to other class classification blocks classified as class information C. Prediction coefficients W ₁ (k) to prediction coefficients W ₂₅ (k) used to calculate the prediction values E [Y ₃₃ (k)] of the pixel value Y ₃₃ (k) (k = 1, 2,..., 9). Once the normal equations that can be computed are established, solve these normal equations to find the optimal prediction coefficient w ₁ suitable for calculating the predicted values E [Y ₃₃ (k)] of pixel value Y ₃₃ (k) for class information C. (k) to the optimum prediction coefficient w ₅ (k) is calculated. The prediction coefficient "w " requires 26, since in this embodiment, 25 training data are used to obtain one prediction value, and one offset coefficient is required. Further, processing operations for establishing normal equations are executed in step S23 until the number of normal equations is obtained. The above processing is performed for each class, and 25x9 prediction coefficients are calculated. Then, prediction values E [Y ₃₃ (k)] are obtained using 25 × 9 prediction coefficients corresponding to 25 pixels in the class information and prediction value calculation block, according to the following equation corresponding to equation (1).

In step S23, after the 25x9 prediction coefficients for each class are calculated, 3x3 prediction values are calculated in units of 3x3 pixels centered on the correction data of interest. Then, in step S24, 25x9 prediction coefficients for each class are provided to the determination unit 24, and prediction values for each unit of 3x3 pixels are provided to the error calculator 23. Thereafter, the process returns to step S21, where, for example, the same processing is repeated for each frame.

Next, FIG. 11 shows a configuration example of the error calculator 23 of FIG. To the block generation unit, raw image data is supplied. In this case, the block generating unit 51 processes the input image data to make nine pixel blocks corresponding to the prediction values output from the local decoder 22. The resulting image blocks (e.g., blocks of 3x3 pixels surrounded by squares in FIG. 7) consisting of 3x3 pixels are supplied from the block generation unit 51 to the squared error calculation unit. As described above, in addition to the blocks supplied from the block generating unit 51, the prediction values derived from the local decoder 22 are squared error calculation unit 52 in units of 9 pixels (3 × 3 pixels in block units). Is supplied. The square error calculation unit 52 calculates a square error as a prediction error of these prediction values for the raw image data, and then supplies this calculated square error to the integrator 55.

In other words, the squared error calculation unit 52 includes calculators 53 and 54. The calculator 53 subtracts corresponding prediction values from each of the block processed image data output from the block generation unit 51, and then supplies this subtraction value to the calculator 54. The calculator 54 squares the output of the calculator 53 (difference between the raw image data and its expected values) and supplies the squared value to the integrator 55.

When the integrator 55 receives the squared error from the squared error calculation unit 52, the integrator 55 reads the stored value of the memory 56, adds the stored value to the squared error, and then adds this addition value to the memory 56. To be stored. Thereafter, the above operation is repeated by the integrator 55, and the multiplication value (error variance) of the squared error is obtained. In addition, when the integration for a predetermined amount of squared errors (for example, the amount of one frame) is completed by the integrator 55, the integrator 55 reads this integration value from the memory 55, and then The integrated value is supplied to the determination unit 24 as error information. Each time the processing operation for one frame ends, the stored value of the memory 56 is cleared, and the output values of the integrator 55 are stored in this memory 56.

Next, the operation of the error calculator 23 is further described with reference to the flowchart shown in FIG. In the error calculator 23, first, in step S31, the stored value of the memory 56 is erased to be, for example, " 0 ". Thereafter, the processing operation proceeds to step S32, in which the image data is block-processed by the block generating unit 51 as described above, and the resulting block is supplied to the square error calculation unit 52. In the square error calculation unit 52, a calculation is performed on the square error between the raw image data constituting the block supplied from the block generation unit 51 and the prediction values supplied from the local decoder in step S33.

That is, in step S33, in the calculator 53, the corresponding predicted value is subtracted from each of the image data blocked and supplied by the block generating unit 51, and then the subtracted value is supplied to the calculator 54. In step S33, the output from the calculator 53 is squared by the calculator 54, and the squared value is supplied to the integrator 55.

When the integrator 55 receives the squared error supplied from the squared error calculating unit 52, in step S34, the squared error is read by adding the stored value of the memory 56 to the squared error. Integrate the value of. The integral value of the squared error integrated in the integrator 55 is supplied to the memory 56 and overwritten over the previously stored value.

Then, in the integrator 55, in step S35, a determination is made as to whether integration of a predetermined amount, for example, squared errors for one frame, has ended. If it is determined in step S35 that the integration of the squared errors for one frame is not finished, the processing operation returns to step S32, and the processing operation defined in step S32 is repeated again. If it is determined in step S35 that the multiplication of the squared errors for one frame is finished, the processing proceeds to step S36. The integrator 55 reads the multiplication value of the squared errors for one frame stored in the memory 56 in step S36, and then outputs this multiplication value to the determination unit 24 as error information. Thereafter, the processing operation returns to step S31, and the processing operations defined in step S31 are repeated.

As a result, in the error calculator 23, assuming that the initial image data is Y _ij (k) and the prediction values processed from the correction data at the local decoder 22 are E [Y _ij (k)], The error information Q can be calculated by performing the calculation according to it.

Q = ∑ (Y _ij (k)-[Y _ij (k)]) ²

Here, the sign "∑" means the sum of one frame of data.

Next, FIG. 13 shows an example of the configuration of the determination unit 24 of FIG.

The prediction coefficient memory 61 stores prediction coefficients supplied from the local decoder 22. The correction data memory 62 stores correction data supplied from the compression unit 21.

When the compressed data is newly corrected, and thus the newly corrected data is supplied, the correction data memory 61 stores the newly corrected data instead of the previously corrected data (that is, the previous correction data). In addition, when such correction data is updated by the newly corrected data, a set of new prediction coefficients for each class corresponding to the newly corrected data is output from the local decoder 22. In addition, in the prediction coefficient memory 61, when a new prediction coefficient for each class is supplied as described above, the new prediction coefficient is the previous prediction coefficient for each class stored in the previous time (i.e., the previous prediction for each class). Coefficients) instead.

The error information memory 63 stores error information supplied from the error calculator 23. In addition to the error information currently supplied, the error information memory 63 also stores error information supplied from the error calculator 23 last time (even if new error information is supplied, until the new error information is supplied again). Stored error information). In addition, each time the processing operation for a new frame is started, the contents of the error information memory 63 are erased.

The comparator 64 may compare the predetermined threshold "ε" with the currently supplied error information stored in the error information memory 63, and if necessary, also compare the currently supplied error information with the previously supplied error information. The comparison result obtained from the comparator 64 is supplied to the controller 65.

The controller 65 determines whether the correction data stored in the correction data memory 62 is exactly (optimally) identical to the coded result of the original image data based on the comparison result obtained from the comparator 64. When the controller 65 recognizes (determines) that the correction data is not optimally equal to the coded result, the controller 65 outputs a control signal that requires the output of new correction data to the compression unit 21 (correction unit 21) (Fig. 5). To feed. On the contrary, if the controller 65 recognizes that the correction data stored in the correction data memory 62 is optimally equal to the coded result of the image, the controller 65 reads the prediction coefficients for each class stored in the prediction coefficient memory 61 The data is output to the multiplexer 25, and the correction data stored in the correction data memory 62 is read out and output to the multiplexer 25 as optimal correction data. Further, in this case, the controller 65 outputs a control signal to the compression unit 21, which indicates a message that the coding operation for one frame has ended. As a result, as described above, the controller 65 causes the compression unit 21 to start the processing operation for the next frame.

Next, the operation of the determination unit 24 will be further described with reference to FIG. 14. In the determination unit 24, the comparator 64 determines whether or not error information has been received from the error calculator 23 in the first step S41. If the comparator 64 determines that the error information has not been received, the processing returns to step S41. On the other hand, if the comparator 64 determines that the error information has been received, that is, if the error information is stored in the error information memory 63, the processing proceeds to step S42. In this step S42, the comparator 64 compares the predetermined threshold "ε" with the currently stored error information (current error information) to determine which one of these values is large.

In this step S42, if it is determined that the current error information is greater than or equal to the predetermined threshold "ε", the comparator 64 reads the previous error information stored in the error information memory 63. Then, in step S43, the comparator 64 compares the previous error information with the current error information to determine whether the previous error information is larger than the current error information.

When the processing operation for one frame image is started and error information is supplied for the first time, since the previous error information is not stored in the error information memory 63, in this case, the determination unit 24 and after step S43 The processing operations defined in will not be performed. The controller 65 outputs a control signal for controlling the correction circuit 32 (Fig. 5) to output a predetermined initial address.

If it is determined in step S43 that the current error information is less than or equal to the previous error information, that is, if the compressed data is corrected and the error information is reduced, the processing proceeds to step S44. In this step S44, the controller 65 outputs a control signal to the correction circuit 32 instructing to change the correction value? Similar to the previous time. Thereafter, the processing returns to step S41. On the contrary, if it is determined in step S43 that the current error information is larger than the previous error information, that is, if the error information is increased by correcting the compressed data, the processing proceeds to step S45. In this step S45, the controller 65 outputs a control signal to the correction circuit 32 which instructs to change the correction value Δ as opposed to the previous time. Thereafter, the processing returns to step S41.

Further, if the continuously reduced error information is increased at a specific timing, the controller 65 outputs a control signal instructing to change the correction value Δ by, for example, 1/2, to change it as opposed to the previous time.

After that, since the processing operations defined in steps S41 to S45 are repeated, the error information is reduced. As a result, when it is determined in step S42 that the current error information is smaller than the predetermined threshold "ε", the processing proceeds to step S46. In this step S46, the control circuit 64 reads the prediction coefficients for each class stored in the prediction coefficient memory 61, and also reads one frame of correction data stored in the correction data memory 62. Then, these optimal compressed data and prediction coefficients for each class are supplied to the multiplexer 25, and the processing operation ends.

Thereafter, waiting for the supply of the error information for the next frame, the processing operations according to the flowchart shown in FIG. 14 are repeated again.

In the correction circuit 32, in order to correct compressed data, all compressed data for one frame can be corrected, and only a part of such compressed data for one frame can be corrected. For example, when only a portion of the compressed data is performed, pixels that can strongly affect the error information are detected by the controller 65, and only the compressed data for such pixels can be corrected. Pixels that can strongly affect the error information can be detected, for example, in the following manner. That is, first, after thinning, a processing operation is performed by directly using the level-control compression data for the remaining pixels, so that such error information is required. Then, such a control signal is supplied from the controller 65 to the correction circuit 32, thereby processing processing for correcting the compressed data for the remaining pixels after thinning by the same correction value Δ for each pixel. Do this. The remaining error information is compared with that error information and obtained when the compressed data is used directly. Next, when the comparison deviation is greater than or equal to a predetermined value, a pixel for error information may be detected as a pixel that may have a strong influence on the error information.

As described above, the correction operation of the compressed data is repeated until the error information is reduced or until it is reduced to be smaller than the predetermined threshold "ε". The corrected data obtained when the error information becomes smaller than the predetermined threshold "ε" is output as a coded result of the picture. Therefore, in the receiving device 4 [FIG. 1], when the pixel value of the pixel for constructing the thinned image is generated as an optimal value used to recover the raw image data, the corrected data (optimal correction data) It is possible to produce a decoded picture (a decoded picture with high quality) that is the same (or substantially the same) as the original picture from.

Since the image is compressed through the thinning processing operation and also by the level limiting processing operation and the class classification adaptive processing operation, it is possible to obtain coded data having a very high compression ratio. Since the compression processing operation by the thinning processing operation is organically combined with the class classification adaptive processing operation in order to realize a highly efficient compression, the above-described coding processing operation performed in the transmission device 1 is referred to as a "combining coding processing operation". It is called.

Referring to Fig. 15, a decoder for use in the first embodiment of the present invention is described. FIG. 15 shows an example of the structure of the receiving device 4 shown in FIG.

In the receiving / reproducing unit 71, coded data recorded on the recording medium 2 is reproduced, or coded data transmitted via the transmission path 3 is received and supplied to the separating unit 72. . In the separating unit 72, the coded data is separated into corrected data and predicted coefficients for each class. The corrected data is supplied to the block unit 73 for class classification and the block unit 77 for calculating the predicted value, and the predicted coefficients for each class are supplied to the prediction unit 76 and next to, for example, a memory (not shown). None).

The class classification block unit 73, the class classification unit 75, or the prediction value calculation block unit 77 may be the class classification block circuit 41, the class classification unit 45, or the prediction value calculation block shown in FIG. It is configured similarly to the generating unit 42. As a result, the predictive value calculation block is output from the predictive value calculation block unit 77 or the class information is output from the class classification unit 75. This prediction value calculation block and class information is supplied to the prediction unit 76.

In the prediction unit 76, 25x9 prediction coefficients are read out from the memory (not shown) in response to class information, and the calculation is 3x according to formula (1) by using read prediction coefficients in response to class information. 3 is composed of 5x5 correction data for constructing the predictive value calculating block supplied from the predictive value and the predictive value calculating block unit 77. Next, the picture for one frame composed of such prediction values is output as a decoded picture. As mentioned above, this decoded picture is substantially the same as the original picture.

Even when the receiving device 4 shown in Fig. 15 is not used at the receiving end, normal interpolation is performed without using the prediction coefficients by the device for decoding the thinned picture by simple interpolation, so that the decoded picture is obtained. It should be noted that it is possible to obtain. In this case, the picture quality (resolution) of the resulting decoded picture is lost.

Second embodiment

In the first embodiment of the present invention, the prediction coefficients are obtained in the local decode 22 of Fig. 3, and the prediction values are then calculated by using these prediction coefficients. Alternatively, each predictive value may be calculated without obtaining the predictive coefficients in local decode 22.

Specifically, the second embodiment of the present invention refers to the transmission device of FIG. 28, the compression unit of FIG. 5, the local decoder of FIG. 16, the error calculator of FIG. 11, the determination unit of FIG. 13, and the learning device for prediction coefficients of FIG. 17. Is explained. A receiving device for use in the second embodiment is shown in FIG. If nothing is mentioned, the compression unit of FIG. 5, the error calculator of FIG. 11, the judging unit of FIG. 13, and the receiving device of FIG. 15 operate as previously described for the first embodiment of the present invention.

Referring to Fig. 28, the transmitting apparatus of the second embodiment is different from that of Fig. 3, wherein the local decoder 22 'of Fig. 28 operates differently from the local decoder 22 of Fig. 3 as described above.

In FIG. 16, FIG. 16 shows another embodiment of the local decoder 22 of FIG. It should be understood that the same reference numerals as in FIG. 8 are used to indicate corresponding circuit parts. In other words, instead of the adaptive processing unit 46, the prediction coefficient ROM 81 and the prediction unit 82 are used in the local decoder 22 'of FIG. The other circuit configuration is similar to that of FIG.

The prediction coefficient ROM 81 stores prediction coefficients for the class obtained by learning (to be described later). When receiving class information output from the class unit 45, the prediction coefficients stored at the address corresponding to this class information are read from the prediction coefficient ROM 81 and then supplied to the prediction unit 82.

In the prediction unit 82, the calculation is performed by the equation (1) (strictly speaking, for example, by using the prediction value calculation block derived from the prediction value calculation block generation unit 42 and the prediction coefficient ROM 81). 8)]. As a result, the predicted value of the raw image data is calculated.

As a result, according to the class classification adaptive processing circuit 43 shown in Fig. 16, this predictive value can be calculated without using the raw image data. The 25x9 prediction coefficients for each class stored in the prediction coefficient ROM 81 are output to the determination unit 24 in the above-described embodiment.

As shown in Fig. 16, when the local decoder 22 'is configured, the receiving device 4 shown in Fig. 15 is configured in this manner so that the front end of the receiving / reproducing unit 71 is the class classification processing circuit (Fig. 16). 43).

With reference to FIG. 29, the operation of the local decoder of FIG. 16 is described in more detail. First, in the local decoder 22 'in step S21, the corrected data derived from the compression unit 21 is processed and blocked. In other words, in the block generation unit 41 for class classification, the correction data is divided into predicted value calculation blocks generated by 5 x 5 pixels located around the corrected data of interest, and the class classification adaptive processing circuit 43 Is supplied.

As described above, the class classification block is supplied to the class classification unit 45 and the prediction calculation block is supplied to the prediction unit 82.

In step S22, the class classification unit 45 performs a class classification processing operation on the class classification block. In other words, the class classification unit 45 detects the state of the distribution of each pixel level in the block, and then determines which class this class classification block belongs to. This class determination result is supplied to the prediction coefficient ROM 81 as class information.

In this embodiment, since the class classification processing operation is performed on a class classification block composed of five pixels represented by two bits, each class classification block is classified into any one of 1024 classes (= (2 ² ) ⁵ ).

Next, the processing operation proceeds to step S23 ', where the prediction coefficients corresponding to each class are read. In step S24 ', the adaptive processing is performed by the prediction unit 82 to calculate the prediction value. In step S25 ', the prediction coefficients for each class are output by the prediction coefficient ROM 81, and the prediction values are output by the prediction unit 82. Then return to step S21 and repeat for example for each one frame.

Next, FIG. 17 shows a structural example of an image processing apparatus for performing a learning operation to receive prediction coefficients stored in the prediction coefficient ROM 81 shown in FIG.

The learning image data (learning image) (hence the image data before thinning) is supplied to the learning block generation unit 91 and the teacher block generation unit 92 used to receive prediction coefficients applicable to any image.

The learning block generation unit 91 for learning extracts, for example, 25 pixels (5x5 pixels) of the positional relationship indicated by the symbol "." Centered on the pixel of interest in FIG. 7, and constituted by such 25 pixels. The block is supplied to the level limiting unit 93 as a kind of learning block. The level limiting circuit 93 limits the level of the input pixel data and outputs the level-limiting pixel data to the class classification unit 94 and the learning data memory 96.

Further, in the teacher block generating unit 92, for example, a block composed of nine pixels (3 x 3 pixels) centered on the pixel of interest is generated from the input image data, and then one block composed of these nine pixels. This teacher block is provided to the teacher data memory 98.

For example, when a learning block composed of 25 pixels is located in the learning learning block generating unit 91 in the positional relationship indicated by the symbol ". &Quot; shown in Fig. 7, the 3x3 enclosed in the rectangular shape shown in Fig. 7 Note that the teacher block of pixels is produced from the teacher block generation unit 92.

The class classification unit 93 extracts the center 9 pixels (3 × 3 pixels) from the 25 pixels to construct the learning block, and then performs class classification on the blocks made of these 9 pixels. The level limiting circuit 93 limits the level of the input pixel data, and outputs the level limiting pixel data to the class classification unit 94 and the learning data memory 96. The resulting class information is supplied to learning data memory 96 and teacher data memory 98 via terminal " a "

In the learning data memory 96 or in the teacher data memory 98, the learning block output from the level limiting circuit 93 or the teacher block output from the teacher block generation unit 92 corresponds to the class information supplied thereto. It is stored at the address AD.

As a result, for example, a block composed of 5x5 shown by the symbol "." In FIG. 7 is stored as a learning block at a certain address of the training data block 96, and is composed of 3x3 pixels indicated in this figure. The block enclosed by the rectangle is stored as a learning block at the same address as that described above.

In succession, similar processing operations are repeated for all learning images provided, so that the learning block and the teacher block are stored at the same address in the learning data memory 96 and the teacher data memory 98. The teacher block is composed of 9 pixels, and the predictive value is calculated in the local decoder 22 of FIG. 16 by using a predictive value calculation block composed of 25 correction data having the same positional relationship as that of the 25 pixels for constructing such a learning block. .

Since a plurality of pieces of information are stored at the same address in the learning data memory 96 and the teacher data memory 98, a plurality of learning blocks and a plurality of teacher blocks can be stored at the same address.

When the learning blocks and teacher blocks for all learning images are stored in the learning data memory 96 and the teacher data memory 98, the switch 95 having the selected terminal "a" is changed from this terminal to the other terminal "b". The output of the counter 97 is supplied to the learning data memory 96 and the teacher data memory 98 as addresses. The counter 97 counts a predetermined clock to output the counter value, and in the learning data memory 96 or the teacher data memory 98, the learning block or teacher block stored at an address corresponding to the counter value is calculated in a calculation unit ( 99) to be supplied.

As a result, in the calculating unit 99, a set of learning blocks and a set of teacher blocks of a class corresponding to the counter values of the counter 97 are supplied.

In the reception of a set of learning blocks and a set of teacher blocks for a certain class, the calculation unit 99 calculates prediction coefficients that can minimize the error in the least squares method by using these block stages.

In other words, for example, the pixel values of the pixels for constructing the learning block are X ₁ , X ₂ , X ₃ , ..., and the required prediction coefficients are w ₁ , w ₂ , w ₃ , ... In this case, the prediction coefficients w ₁ , w ₂ , w ₃ , ... must satisfy the following equation to obtain the pixel value "y" of a constant pixel for constructing the teacher block by the linear equation of these values. do.

y = w ₁ x ₁ + w ₂ x ₂ + w ₃ x ₃ + ... + w ₀

As a result, the prediction coefficients w ₁ , w ₂ , w ₃ , ..., w ₀ are calculated in the calculation unit 99 in this manner, so that the normal equation indicated in the above-described formula (7) is set and solved. This predictive coefficient is the predicted value (w ₁ x ₁ + w ₂ x ₂ + w ₃ x ₃ + ... + w ₀ ) for the true value "y" from both the learning block and the corresponding teacher block in the same class. Minimize the square error of Therefore, the 25x9 prediction coefficient for each class is calculated by performing the above-described processing for each class.

The prediction coefficients for each class calculated in the calculation unit 99 are supplied to the memory 100. The calculated values are supplied from the counter 97 into the memory 100 in addition to the prediction coefficients derived from the calculation unit 99. As a result, in the memory 100, the prediction coefficients derived from the calculation unit 99 are stored at an address corresponding to the calculated value output from the counter 97.

As mentioned above, the optimal 25x9 prediction coefficients used to predict the 3x3 pixels of the blocks in this class are stored at the addresses corresponding to each class in memory 100.

The prediction coefficients for each class stored in the memory 100 of the above-described manner are stored in the prediction coefficient ROM 81 of FIG. In the second embodiment, since the prediction coefficients for each class are not calculated in the local decoder 22 'of Fig. 16, it is possible to increase the processing speed.

As shown in Fig. 16, in the configuration of the local decoder 22 ', the receiving device 4 shown in Fig. 15 is configured in such a manner so that the rear end of the receiving / reproducing unit 71 is the class classification processing of Fig. 16. It is configured similarly to the circuit 43.

Third embodiment

In the first embodiment of the present invention, the prediction coefficients are accommodated in the local decoder 22 of Fig. 3, and the prediction values are calculated by using these prediction coefficients. Alternatively, each prediction coefficient does not need to be calculated, but instead transmission and / or recording may be omitted and then calculated in the receiving device.

Specifically, the third embodiment of the present invention is described with reference to the transmission device of FIG. 30, the compression unit of FIG. 5, the local decoder of FIG. 32, the error calculator of FIG. 11, and the determination unit of FIG. A receiving device for use in this third embodiment is shown in FIG. If nothing is mentioned, the compression unit of FIG. 5 and the error calculator of FIG. 11 operate as previously described in the first embodiment of the present invention.

Referring to FIG. 30, the transmitting apparatus of the third embodiment is different from that of FIG. 3 in that the local decoder 22 'of FIG. 28 operates differently from the local decoder of FIG. Further, the determination unit 24 'of FIG. 30 does not output the prediction coefficients for each class, and therefore, the multiplexer 25 as in FIGS. 3 and 28 is not necessary.

Referring to FIG. 31, the operation of the transmission device of FIG. 30 is described. When the image data is supplied to the compression unit 21, the compression unit 21 first compresses the image data in step S1 and thins the image data (that is, a plurality of pixels from a plurality of pixels. Extracted]. In step S2, the level of the extracted pixel data is limited by extracting two bits from such pixel data on the MSB side. Next, such pixel data is initially output to the local decoder 22 'and the determination unit 24 without any correction. In the local decoder 22 'in step S3, the correction data derived from the compression unit 21 is locally decoded. This corrected data is the same as such compressed data, and as described above, the image data is simply thinned and also initially level-controlled.

In step S3, the predicted value is calculated. The predicted value calculated from the local decoder 22 'is supplied to the error calculator 23.

In this case, the picture composed of the predicted values output from the local decoder 22 '' is the same as the decoded picture (Fig. 1) accepted on the receiving device 4 side.

In step S4, when the predicted value of the raw image data is received from the local decoder 22 '' by the error calculator 23 in step S4, the error calculator 23 performs the local decoder 22 '' on the raw image data. The prediction error of the prediction value derived from is calculated, and this prediction error is supplied to the determination unit 24 'as error information. When the error information is received from the error calculator 23, the determination unit 24 'determines in step S5 whether the corrected data output from the compression unit 21 exactly matches the coded result of the original image.

In other words, in step S5, for example, it is determined whether the error information is smaller than the predetermined threshold value "?". In step S5, if the error information is not smaller than the predetermined threshold "ε", it is recognized that the corrected data output from the compression unit 21 is not the same as the coded data of the original picture. Next, the processing operation proceeds to step S6 in which the determination unit 24 'controls the compression unit 21 to correct the compressed data. The compression unit 21 changes the correction amount under the control of the determination unit 24 'to correct the data (correction value " Δ " will be described later). The resultant corrected data is the local decoder 22' and the determination unit ( 24 ') Next, the processing returns to the previous step S3, and a similar processing operation is repeatedly performed.

On the other hand, in step S5, when it is determined that the error information is smaller than the predetermined threshold "ε", it is recognized that the corrected data output from the compression unit 21 is exactly the same as the coded result of the raw image data. When the error information is smaller than the predetermined threshold "ε", the determination unit 24 outputs the optimum compressed data. Next, the process ends.

As described above, since the corrected data is generated by correcting the compressed data, the condition under which the error information becomes smaller than the selected threshold value "ε" is used as the coded result of the raw image data, and this reference data on the receiving device side is referred to. It is possible to accommodate an image which is actually identical to the raw image data.

Referring to FIG. 32, a block diagram of a local decoder 22 'is shown. In contrast to the local decoder 22 'of Fig. 16 (of the second embodiment), the local decoder 22' does not output the prediction coefficients for each class from the prediction coefficient ROM 81 '.

Referring to Fig. 33, the operation of the local decoder 22 'in Fig. 32 is described. First, in the local decoder 22 'in step S21, the corrected data derived from the compression unit 21 is processed to generate a block. In other words, in the class classification block generation unit 41, the correction data is divided into a class classification block composed of five pixels centered on the noted correction data, which is supplied to the processing circuit 43 centered on the class classification. do. In the predictive value calculation block generating unit 42, the corrected data is divided into a predictive value calculation block composed of 5x5 pixels centered on the corrected correction data, and supplied to the class classification adaptive processing circuit 43.

As described above, the class classification block is supplied to the class classification unit 45 and the prediction calculation block is supplied to the adaptive processing unit 46.

In step S22, the class classification unit 45 performs class classification processing on the class classification block. In other words, the class classification unit 45 detects the distribution state of each pixel level in the block, and then determines which class belongs to this class classification block. This class determination is supplied to the prediction coefficient ROM 81 as class information.

In this embodiment, since the class classification processing operation is performed on a class classification block composed of five pixels represented by two bits, each class classification block is classified into one of 1024 classes (= (2 ² ) ⁵ ).

Next, the processing operation proceeds to step S23 ', and prediction coefficients corresponding to each class are read. In step 24 ', adaptive processing is performed by the prediction unit 82 to calculate the prediction value. In step S25 '', the predictive value is output by the prediction unit 82. Step S21 returns to the next, and the same processing is repeated, for example, for each one frame.

Referring to Fig. 34, the determination unit 24 'of the third embodiment is described. In contrast to the determination units of the first and second embodiments as shown in FIG. 13, the controller 65 ′ of FIG. 34 does not output prediction coefficients for each class. In addition, the determination unit 24 'does not include the prediction coefficient memory shown in 61 of FIG.

Referring to Fig. 35, the operation of the determination unit 24 'of Fig. 34 is further described. In the judging unit 24 'for the judging unit in FIG. 13 in the first step S41, the comparator 64 determines whether or not the error information is received from the error calculator 23. FIG. If the comparator 64 determines that the error information is not received, the processing operation returns to step S41. In other words, when the comparator 64 determines that the error information has been received, in other words, when the error information is stored in the error information memory 63, the processing proceeds to step S42. In this step S42, the comparator 64 compares the currently stored error information (current error information) with a value equal to or greater than the predetermined threshold "ε" to determine whether any of these values is large.

When the processing operation for one frame image is started and the error information is supplied first, since the previous error information is not stored in the error information memory 63, in this case, the processing operation limited after step S43 is determined by the determination unit 24. Is not performed within. The controller 65 outputs a control signal for controlling the correction circuit 32 (Fig. 5) to output a predetermined initial address.

In step S43, when the current error information is less than or equal to the previous error information, that is, when the compressed data is corrected to reduce the error information, the processing proceeds to step S44. In step S44, the controller 65 outputs a control signal to the correction circuit 32 for instructing that the correction value [Delta] is changed in a manner similar to that in the previous case. Next, the processing operation returns to step S41. In contrast, if the current error information is larger than the previous error information in step S4, that is, if the error information is increased by correcting the compressed data, the processing proceeds to step S45. In step S45, the controller 65 outputs a control signal to the correction circuit 32 for a command that the correction value [Delta] changes in the opposite manner to the previous case. Next, the processing returns to step S41.

In the case where the continuously reduced error information increases at a certain time, the controller 65 outputs a control signal for the command that the correction value Δ is changed in the opposite manner as in the previous case. It should also be noted that it outputs a control signal for a command that decreases by two and changes in the opposite way to the previous case.

Next, the limited processing operation is repeated from step S41 to step S45, and the error information is reduced. As a result, when it is determined in step S42 that the current error information is smaller than the predetermined threshold value "ε", the processing proceeds to step S46. In this step S46, the control circuit 64 reads out one frame of corrected data stored in the corrected data memory 62. Therefore, the optimal captured data is supplied to the transmission / playback apparatus, and the process ends.

From this, the processing operation is repeated again according to the flowchart shown in Fig. 34, and waits to supply error information for the next frame.

Referring to Fig. 36, the reception device 4 'used in the third embodiment is described. In contrast to the receiving apparatuses (FIG. 15) of the first and second embodiments, since the prediction coefficients for each class are not provided to the receiving / reproducing unit 71, a separate unit 72 as in FIG. 15 is not necessary. not. In addition, since the prediction coefficients for each class are not provided to the receiving / reproducing unit 71, the prediction coefficient ROM 78 is used to provide the prediction coefficients for each class to the prediction unit 76. In other words, the operation of the receiving device 4 'of FIG. 36 is the same as that of the receiving device 4 of FIG.

It is to be understood that the prediction coefficients are not stored at addresses corresponding to the respective classes in the prediction coefficients ROM 81 and 78, but the average value of the pixel values for constructing the teacher block can be stored. In this case, when class information is supplied, pixel values corresponding to this class are output. Thus, in the local decoder 22 of FIG. 16, the predictive value calculating block generation unit 42 and the prediction unit 82 are not provided.

Although the sum of squares of errors is used as the error information of this embodiment, other factors such as the sum of the absolute value of the error sum and the sum of the cubic errors can be used as the error information. For example, a determination is made as to which error information is used based on the convergence characteristic.

Further, in this embodiment, the correction operation of the compressed data is repeated until the error information becomes smaller than the prediction threshold "ε". Alternatively, it is also possible to generate an upper limit on the number of correction operations on the compressed data. That is, when the image is transmitted in real time, the processing operation for one frame must be finished within a predetermined time period. There is no explicit condition that error information converges within such time period. Therefore, since the upper limit value includes the number of corrections, if the error information does not converge to a value smaller than the threshold value of? In a predetermined time period, the processing operation for such a frame is completed (at the same time, the corrected data is coded). Recognized as a result], the processing operation for the next frame is started.

Also, in this embodiment, the block is composed of images for one frame. Optionally, the block may consist of pixels at the same location during multiple time-order frames.

Further, in this embodiment, the image is simply thinned in the compression unit 21, that is, the center pixel in the block composed of 3x3 pixels is extracted, which is then used as compressed data. Optionally, the compression unit 21 reduces the number of pixels (thinning) by calculating an average value of nine pixels for constituting the block and setting this average value as the pixel value of the center pixel in the block, thereby reducing this reduced data. Can be used as compressed data.

Fourth embodiment

A fourth embodiment of the present invention will now be described with reference to the transmission device of FIG. 18, the learning device for the mapping coefficients of FIG. 21, the local decoder of FIG. 23, and the learning device for the prediction coefficients of FIG. 17. The receiving device used in this fourth embodiment is shown in FIG. Except as otherwise indicated, the learning apparatus for the prediction coefficients of FIG. 17 operates as described above in connection with the second embodiment of the present invention.

Referring now to Fig. 18, the transmitting apparatus of the fourth embodiment is described.

The coded picture data is provided to the block generation unit 111.

The block generating unit 111 processes the image data, forms a class classification block centered on the indicated pixel corresponding to the unit classifying the image data into a predetermined class according to the attribute thereof, and then classifies the class classification block. To the ADRC (Adaptive Dynamic Range Coding) processing unit 112 and the delay unit 115.

The ADRC processing unit 112 executes an ADRC processing operation on the block (class classification block) obtained from the block generation unit 111, and provides a block composed of the final ADRC code to the class classification unit 113.

In this case, by the ADRC processing operation, the number of bits of pixels for constituting the class classification block can be reduced.

That is, for the sake of simplicity, for example, as shown in Fig. 19 (a), when a block composed of four pixels arranged in a straight line is considered, the maximum value " MAX " and minimum value " MIN " Is detected in the ADRC processing operation. And now consider a case where DR = MAX-MIN is used as the local dynamic range of the block, the pixel value of the pixels constituting the block quantizes k bits again based on the dynamic range DR.

That is, the minimum value MIN is the basis of each pixel value in the block, and the default value is divided by DR / 2 ^k . Therefore, the final value is converted into a code (ADRC code) corresponding to the divided value. Specifically, for example, as shown in Fig. 19B, when k = 2, any one such range obtained by dividing the dynamic range DR equally by 4 (2 ² ) It is determined whether or not belong to. When the division value is included in the lowest level range, the second lowest level range, the third lowest level range, or the highest level range, each division value is 00B, 01B, 10B, or 11B (symbol "B" is a binary number). Coded as a bit). On the decoding side, the ADRC code 00B, 01B, 10B, or 11B is the median value L00 of the lowest level range, the median value L01 of the second lowest level range, the third lowest level median L10, or the highest range level. Is converted to the median L11 of the range. The minimum value MIN adds each of these values for the decoding operation to be performed.

In this case, such ADRC processing operation is referred to as "non-edge matching". As pointed out in Fig. 19 (c), the case opposite to the non-edge matching. It is referred to as an enhanced "non-edge match" corresponding to the ADRC processing operation. That is, in the ADRC processing operation, one of ADRC code 00B or ADRC code 11B divides the dynamic range DR into four equally, so that the average value MIN of pixel values included in the lowest level dynamic range obtained by the same or the highest value. It is converted into another average value MAX of the pixel values included in the level range. Also, the ADRC codes 01B and 10B are converted into levels obtained by equally dividing the dynamic range DR defined by (MAX'-MIN ') * 3 for decoding the ADRC code.

The detailed description relating to such an ADRC processing operation is described in detail in, for example, Japanese Laid-Open Patent Application 3-53778, already filed and published in 1991.

As already explained, the number of classes can be reduced by performing the ADRC processing operation as the requantization is performed based on the number of bits smaller than the number of bits allocated to the pixels for constructing the block. Returning to the second half of Fig. 18, the class classification unit 113 performs a class classification processing operation for classifying a block output from the ADRC processing unit 112 into a predetermined class, depending on such an attribute, and this block Such information is supplied to the mapping coefficient memory 114 to indicate that it contains any class as this classification information.

For example, when eight or more bits are allocated to each pixel, the class classification block is arranged by nine pixels 3x3 in the center of the marked pixel. The class classification processing operation is performed in connection with such a class classification block, which is classified into a huge number of classes, for example, (2 ⁸ ) ⁹ classes.

Therefore, according to this embodiment, the ARDC processing operation is performed on the class classification block in the ARDC processing unit 112. Therefore, the number of classes is reduced as the number of bits of pixels for forming the class classification block is increased, and the number of classes is further reduced. That is, for example, when one bit of the ADRC processing operation is performed in the ADRC processing unit 112, the class number is reduced from (2 ⁸ ) ⁹ to (2 ¹ ) ^9, ie 512.

It can also be pointed out that the class classification processing operation indicated in connection with the present embodiment is performed in the class classification unit 113 based on the ADRC code output from the ADRC processing unit 112. Such class classification processing operations may include, for example, predictive decording (DPCM), block truncation cording (BTC), vector quantization (VQ), discrete cosine conversion (DCT), and the like. And data processed by Hadamard conversion, and the like.

18, mapping coefficients obtained from the learning operation (mapping coefficient learning) of all classification information are stored in the mapping coefficient memory 114 as described later. While using all the classification information supplied from the class classification unit 113, the mapping coefficient stored at this address is read out from this memory 111 to be supplied to the calculation unit 116.

The delay unit 115 delays the block supplied from the block generation unit 111 until the mapping coefficient corresponding to the classification information of this block is read from the mapping coefficient memory 114, and the calculation unit ( 116) the delayed block.

The calculation unit 116 is supplied from the delay unit 115 to calculate such coded data, obtained by not dense (or reducing) the number of pixels of this picture coding this picture. A predetermined calculation is performed using the pixel values of the pixels constituting the block and the mapping coefficients corresponding to the class of these blocks, also supplied from the mapping coefficient memory 114. That is, the calculation unit 116 calculates a preselected function value "f" (y ₁ , y ₂ , ..., k ₁ , k ₂ , ...) using the following value as an independent variable. That is, in order to construct a block output from the block generation unit 111, the pixel value of each pixel is y ₁ , y ₂ ,... And the mapping coefficients corresponding to the class of these blocks output from the mapping coefficient memory 114 are k ₁ , k ₂ ,... to be. Such a calculated function value "f" (y ₁ , y ₂ , ..., k ₁ , k ₂ , ...) constitutes, for example, the block (class classification block) output from the block generation unit 111. Is output as a pixel value of a center pixel among the pixels.

Therefore, considering the case where the number of pixels for constructing the class classification block output from the block generation unit 111 is equal to the number of N pixels, the calculation unit 116 reduces image data by 1 / N and The data reduced to 1 / N is output as code data.

The coded data output from the calculating unit 116 is not only data calculated to be selected and calculated for outputting the center pixel of the block arranged by N pixels, apparently simply to execute a reduced processing operation. As already described, it can be seen that the function value "f" (y ₁ , y ₂ ,..., K ₁ , k ₂ ,...) Defined by the N pixels constituting such a block is the same. Looking at it from a different angle, this function value "f" (y ₁ , y ₂ , ..., k ₁ , k ₂ , ...) may be the optimal value. That is, the pixel value of the center pixel in the block, obtained from the simple reduction processing operation, is corrected to the optimum value from which original image data can be obtained based on the pixel value located around this center pixel. Thus, if necessary, the data obtained by calculating the mapping coefficient and the pixels constituting the block may also be referred to as "correction data".

Further, the calculation processing performed in the calculation unit 116 is performed by the function " f " (y ₁ , y ₂₎ on the pixel value of each pixel constituting the class classification block output by the block generation unit 111. It can be thought of as a processing operation for mapping..., K ₁ , k ₂ ,. Therefore, the coefficients k ₁ , k ₂ ,... Will be referred to as the "mapping coefficient". The calculating unit 116 supplies the code data to the level limiting unit 117.

The level limiting unit 117 extracts 2-bit data on the MSB side of 8-bit coded data and supplies the coded data, limited to 2-bit data, to the transmission / recording device 118.

The transmission / recording apparatus 118 records correction data supplied as coded data from the level limiting unit 117 on the recording medium 2 or transmits this correction data via the transmission path 3. .

Referring now to the flowchart of FIG. 20, the operation of the transmitting device of FIG. 18 is further described.

The image data is supplied to the block generating unit 111 in units of one frame (field), for example. In step S61, in the block generation unit 111, one frame of images is processed to form a class classification block. That is, for example, as shown in FIG. 7 enclosed by a rectangle, the block generating unit 111 has nine pixels (3 * 3 pixels along the horizontal and longitudinal directions) in the center of the pixel pointed to the image data. The class classification block is subdivided into a class classification block constituted by the above, and the class classification block is continuously supplied to the ADRC processing unit 112 and the delay unit 115.

In this case, the class classification block is arranged in a square formed of 3x3 pixels. Alternatively, the shape of this class classification block is not limited to a rectangle, but may be formed in, for example, rectangular, cross and any other shape. In addition, the total number of pixels constituting the class classification block is not limited to only nine pixels (that is, 3x3 pixels). Therefore, the class classification block is not arranged by adjacent pixels, but may be arranged in other ways by such pixels being separated from each other. Such shape and pixel amount of the classification class block must match the shape and amount during performing the learning operation (mapping coefficient learning operation will be described later).

In step S62, upon receipt of the class classification block from the block generation unit 111, the ADRC processing unit 112 associates with this block in order to form a block composed of, for example, pixels represented by 1 bit. Performs the ADRC processing operation of the bit. The class classification block that has been ADRC processed is provided from the class classification unit 113.

In step S63, in the class classification unit 113, the class classification block provided from the ADRC processing unit 112 is classified into a class, and the final classification information is supplied to the mapping coefficient memory 114 as an address. Accordingly, mapping coefficients corresponding to the classification information provided from the class classification unit 113 are read out from the mapping coefficient memory 114 and supplied to the calculation unit 116.

On the other hand, while the class classification block provided from the block generation unit 111 is delayed in the delay unit 115, the mapping coefficient corresponding to the classification information of the block is waited to be read from the mapping coefficient memory. The delayed class classification block is supplied. In step S64, in the calculator 116, the function " f (·) " (the symbol "" indicated in this function) is the set of pixel values y ₁ , y ₂ ,... And the mapping coefficient k ₁ , k ₂ ,. Is calculated using the pixel value of each pixel and the mapping coefficient read from the mapping coefficient memory 114 to construct a class classification block provided from the delay unit 115. Therefore, it is possible to calculate such correction data calculated by correcting the pixel value of the central pixel to construct the class classification block. Such correction data is output to the level limiting unit 117 as the code image data.

In step S65, the 8-bit coded data of the level limiting unit 117 is supplied to the transmission / recording device 118 as 2-bit data on its MSB side. In step S66, the transmission / recording device 118 records the coded data supplied from the level limiting unit 117 on the recording medium 2 or transmits it via the transmission medium 3.

Then, before step S67, it is determined whether or not the image data processing operation for one frame has been performed. In this step S67, when it is determined that the image data processing operation for one frame has not yet been performed, the processing operation returns to the previous step S62, and the processing operation determined after the step S62 is repeated for the next class classification block. do. On the contrary, if it is determined in step S67 that the processing operation on the image data for one frame has been performed, the processing operation returns to S61 to repeat the processing operation defined after step S61 for the next frame.

Next, FIG. 21 is a block diagram of an image processing apparatus for performing a learning (mapping coefficient learning) processing operation used to calculate the mapping coefficient stored in the mapping coefficient memory 114 shown in FIG. 18. .

In the memory 121, digital image data for training of one or more frames suitable for the learning process (if necessary, will be referred to as "learning image" below) is stored. The block generating unit 122 reads the image data stored in the memory 121, is composed of the same class classification block as that output from the block generating unit 111 shown in FIG. 18, and processes the block by ADRC processing. Supply to the unit 123 and the calculation unit 125.

Any one of the ADRC processing unit 123 or the class classification block 124 performs a processing operation similarly to the ADRC processing unit 112 or the class classification block 113 shown in FIG. 18. Accordingly, classification information of the block output from the block generation unit 122 is output from the class classification unit 124. This classification information is supplied to the mapping coefficient memory 131 as an address.

The calculation unit 125 calculates the same as that of the calculation unit 116 of FIG. 18 by using the pixels for constructing the block supplied from the block generation unit 122 and the mapping coefficient supplied from the mapping coefficient memory 131. Do this. Therefore, the calculation unit 125 provides the compressed data to the level limiting unit 126. This level limiting unit 126 limits the pixel level of the compressed data and supplies the final correction data (function “f (·)”) to the local decoder 127.

On the basis of the correction data supplied from the level limiting unit 126, the local decoder 127 is configured to predict pixels of the original learning image data (obviously, pixels constituting blocks output from the block generating unit 122). Predicts (calculates) the pixel value of < RTI ID = 0.0 >, < / RTI > The error calculator 128 reads the pixel value (truth value) of the training image corresponding to the prediction value supplied from the local decoder 127, and calculates a prediction error of the prediction value corresponding to the pixel value of the training image. (Detected) and supply this calculated prediction error to the determination unit 129 as error information.

The determination unit 129 compares the error information supplied from the error calculator 128 with a predetermined threshold value "ε1" and controls the mapping coefficient setting unit 130 according to the comparison result. Under the control of the determining unit 129, the mapping coefficient setting unit 130 sets (converts) a set of mapping coefficients, a number corresponding to the number of classes obtained by the class classification unit 124 (transform), and a mapping coefficient memory. Supply a transformed set of mapping coefficients to 131.

The mapping coefficient memory 131 temporarily stores the mapping coefficient supplied from the mapping coefficient setting unit 130. The mapping coefficient memory 131 may include a storage area capable of storing mapping coefficients (a set of mapping coefficients) and the number of the areas is equal to the number of classes classified in the class classification unit 124. In each storage area, when a new mapping coefficient is supplied from the mapping coefficient setting device 130, this new mapping coefficient may be stored in place of the previously stored mapping coefficient.

In addition, the mapping coefficient memory 131 reads the mapping coefficient stored at an address corresponding to the classification information supplied from the class classification unit 124, and supplies the read mapping coefficient to the calculation unit 126.

Referring to the flowchart of FIG. 22, the operation of learning the mapping coefficients is further described.

First, in step S71, the mapping coefficient setting unit 130 sets initial values of a set of coefficients only in the same number as the classes classified in the class classification unit 124, and then the set of initial values described above. Is supplied to the mapping coefficient memory 131. In the mapping coefficient memory 131 described above, mapping coefficients (initial values) derived from the mapping coefficient setting unit 130 are stored in addresses of the corresponding classes.

Then, in step S72, the block generation unit 122 processes all the learning images stored in the memory 121 to process the recorded pixels in a manner similar to that executed in the block generation unit 111 shown in FIG. Create a block of 3 x 3 pixels in the center. In addition, the block generating unit 122 reads the block from the memory 121 and sequentially supplies the read block to the ADRC processing unit 123 and the calculating unit 125.

In step S73, similar to the processing operation by the ADRC processing unit 112 of FIG. 18, the ADRC processing unit 123 performs an 1-bit ADRC processing operation on the block output from the block generation unit 122. This ADRC processed block is supplied to the class classification unit 124.

In step S124, the determination of the class classification unit 124 is made of the class of the block supplied from the ADRC processing unit 123, so that the class information is supplied to the mapping coefficient memory 131 as an address. As a result, in step S75, the mapping coefficients are read from the address of the mapping coefficient memory 131 corresponding to the class information supplied from the class classification unit 124, so that the read mapping coefficients are read to the calculation unit 126. Supplied.

When a block is received from the block generation unit 122 and a mapping coefficient corresponding to the class of this block is received from the mapping coefficient memory 131, the calculation unit 125 performs the above mapping coefficients and the block generation unit 122. The above-described function value “f (·)” used by the pixel values of the pixels constituting the block supplied from is calculated. This calculation result is supplied to the level limiting unit 126 as compressed data obtained by correcting the pixel value of the center pixel of the block supplied from the block generating unit 122. Then, as a result, in step S77, 2-bit data from the MSB side is selected from 8-bit data so that pixel data of bit-limited 2-bit data is supplied to the local decoder 127.

That is, for example, in FIG. 7, assuming that a block of 3 x 3 pixels surrounded by rectangles is output from the block generation unit 122, the calculation unit 125 determines that the pixels indicated by the symbol " The correction data generated by correcting the pixel values is calculated, and the correction data is level limited by the level limiting unit 126, and outputs this calculated correction data to the local decoder 127.

As a result, in the calculation unit 126, the number of pixels constituting the training image data is reduced to 1/9, which is supplied to the local decoder 127.

Returning to Fig. 22, the processing operation proceeds to step S78, where it is checked whether correction data for all learning images stored in the memory 121 has been calculated. When the correction data for all the learning data have not yet been calculated in step S78, the processing operation returns to the previous step S73. Then, processing operations from step S73 to step S78 defined after step S73 are repeated until the correction data for all the learning images is calculated.

When the correction data for all the learning pictures is calculated in step S78, that is, when the thinning picture generated by reducing all the learning pictures stored in the memory by 1/9 can be obtained, the processing operation is performed locally. The decoder 127 proceeds to step S79 in which the picture reduced by 1/9 is locally decoded, and the predicted values of the original learning picture data are calculated. The image thinned to 1/9 is not the same as the thinned image generated by simply thinning the learning image data up to 1/9, yet another obtained by calculating the mapping coefficient to obtain pixel values. Same as thinning image. The above prediction values are then fed to the error calculator 128.

In the above case, the picture constituted by the prediction values obtained at the local decoder 127 (picture obtained when the error information output from the error calculator 128 becomes smaller than the threshold value epsilon 1 as described later) is obtained. It is the same as the decoded picture obtained on the receiving device 4 (FIG. 1) side.

In the error calculator 128, the learned picture is read from the memory 121 in step S80, and then the prediction error of the prediction values supplied from the local decoder 127 for the read data is calculated. That is, when the pixel value of the training image data is represented by "Y _ij ", and the predicted values output from the local decoder 127 are represented by "E [Y _ij ]", the error represented by "Q" by the following equation: The variance (ie, sum of squares of errors) is calculated by the error calculator 128. The error variance value described above is supplied to the determination unit 129 as error information.

, 여기서 기호 "Σ"는 학습 화상들의 모든 화소들의 합을 표시한다.

, Where the symbol " Σ " represents the sum of all pixels of the learning images.

During the reception of the error information from the error calculator 128, the determination unit 129 compares the error information with the previously selected threshold value " ε1 ", so that the above error information is the threshold value " ε1 " in step S81. Determine if greater than When it is determined in step S81 that the error information is greater than or equal to the threshold "ε1", that is, when the picture composed of the prediction values obtained in the local decoder 127 is not recognized as the same as the original learning picture data, the determination unit 129 outputs a control signal to the mapping coefficient setting unit 130. In step S82, in response to the control signal derived from the determination unit 129, the mapping coefficient setting unit 130 changes the mapping coefficient, so that the changed mapping coefficient is newly stored in the mapping coefficient memory 131. .

Then, the processing operation returns to the previous step S73, where the processing operations defined after the step S73 are repeated using the changed mapping coefficients stored in the mapping coefficient memory 131.

It will be appreciated that the mapping coefficient may be changed randomly in the mapping coefficient setting unit 130. In contrast, when the current error information becomes larger than the preceding error information, the mapping coefficient may change according to a trend similar to the preceding trend. In addition, when the current error information is larger than the preceding error information, the mapping coefficient may change according to a trend opposite to the previous trend.

In addition, the mapping coefficient can be changed according to all classes or parts of classes. When the mapping coefficient is changed according to the portion of the classes, for example, a class that can severely affect the error information is detected, and then the mapping coefficient can be changed. Classes that can severely affect error information may be detected as follows. That is, first of all, the processing operation is performed by using the initial value of the mapping coefficient to obtain error information. Thereafter, the mapping coefficient is changed by the same amount per class, and the result error information is compared with the error information obtained when the initial value is used. Classes where the comparison difference is greater than or equal to a predetermined value may be detected as a class having a severe influence on the error information.

Further, in the case where a plurality of mapping coefficients such as k ₁ , k ₂ , ... described above are set as one set, only the mapping coefficients that severely affect the error information can be changed.

Moreover, in the above case, the mapping efficiency is set for each class. Alternatively, the mapping efficiency can be set, for example, regardless of the units of each block, or adjacent blocks.

For example, when mapping coefficients are set regardless of block, there are several possibilities that multiple sets of mapping coefficients are obtained for a given single class (inversely, if a mapping coefficient is not obtained, the class There are several possibilities that arise). Finally, the mapping coefficients are determined for each class as already described, so that when a plurality of sets of mapping coefficients are obtained according to a given class, a predetermined processing operation, a combination operation or selection operation is applied to the plurality of sets of mapping coefficients. One set of mapping coefficients must be determined to determine.

On the other hand, when it is determined in step S81 that the error information is smaller than the threshold "ε1", that is, when it is recognized that the picture composed of the prediction values obtained by the local decoder 127 is the same as the original learning picture data, the processing operation is performed. Is done.

At this time, the mapping coefficients for each class, which are stored in the mapping coefficient memory 131, are the mappings of FIG. 18 as the optimum mapping coefficients that enable to obtain correction data from which a decoded picture that can be recognized as the same picture as the original picture can be generated. The coefficient memory 114 is set.

As a result, since the correction data is generated by using the above mapping coefficients, it becomes possible to obtain an image substantially the same as the original image data on the receiving device 4 (Fig. 1) side.

As described above, in the embodiment of FIG. 21, since the 1 bit ADRC processing operation is performed on the image block in the ADRC processing unit 123, the image is displayed by the pixel (3 x 3) displayed by the block generation unit 122. Subdivided into blocks of 9 pixels in the center of the pixels). As a result, the number of classes obtained by class classification by the class classification unit 124 is 512 (= (2 ¹ ) ⁹ ). Thus, 512 sets of mapping coefficients are obtained.

Next, FIG. 23 shows an example of the arrangement of the local decoder 127 shown in FIG.

The correction data derived from the level limiting unit 126 of FIG. 21 is supplied to the block generating unit 141 for class classification and the block generating unit 142 for calculating the predicted value. The block generating unit 141 for class classification locates the class classification blocks at the center of the displayed corrected data in the unit used to classify the corrected data according to the characteristic into a preselected class.

That is, for example, correction data (compression data), i.e., pixels (indicated by the symbol & quot _;. &Quot; in FIG. 7) at the i-th position from the top and the j-th position from the left in FIG. Assuming that it is expressed, the block generating unit 141 for class classification includes the target pixel "X _ij " and corresponding upper, left, right and lower positions X _(i-1) , X _{i (j-1)} , X It will construct a class classification block arranged in a total of 5 pixels, i.e. _{, j ( 1 + 1)} and 4 pixels adjacent to the target pixel, X _{(i-1) j} .

It will be appreciated that the class classification block obtained in the block generation unit 141 for class classification in FIG. 23 is configured to determine the class of the block used for calculating the prediction values. In this respect, the class classification block described above is different from the class classification block generated in the block generation unit 111 in FIG. 18 to determine the class of the block in order to calculate the correction data.

The block generation unit 142 for calculation of the prediction values processes the correction data to produce a prediction value calculation block at the center of the displayed correction data used to calculate the prediction values of the raw image data (ie, the learning image data in this case). Occurs. That is, in this embodiment, the pixel values of the nine pixels (= 3x3) of the raw image data at the center around the correction data X _ij are Y _ij (1), Y _ij (2), Y _ij (3), Y _{It is represented by ij} (4), Y _ij (5), Y _ij (6), Y _ij (7), Y _ij (8), and Y _ij (9). The above pixel values are obtained from 25 pixels (= 5x5) at the center around the pixel "X _ij ". These 25 pixels are X _{(i-2) (j-2)} , X _{(i-2) (j-1)} , X _{(i-2) j} , X _{(i-2) (j + 1)} , X _{( i-2) (j + 2)} , X _{(i-1) (j-2)} , X _{(i-1) (j-1)} , X _{(i-1) j} , X _{(i-1) (j +1)} , X _{(i-1) (j + 2)} , X _{i (j-2)} , X _{i (j-1)} , X _ij , X _{i (j + 1)} , X _{i (j + 2)} , X _{(i + 1) (j-2)} , X _{(i + 1) (j-2)} , X _{(i + 1) (j-1)} , X _{(i + 1) j} , X _{(i + 1 ) (j + 1)} , X _{(i + 1) (j + 2)} , X _{(i + 2) (j-2)} , X _{(i + 2) (j-1)} , X _{(i + 2) j} , X _{(i + 2) (j + 1)} , X _{(i + 2) (j + 2)} .

Specifically, for example, in FIG. 7, in order to calculate the prediction values of the pixels Y33 (1) to Y33 (9) of the raw image data surrounded by the rectangles, the predictive value calculation block is X ₁₁ , X ₁₂ , X _13. , X ₁₄ , X ₁₅ , X ₂₁ , X ₂₂ , X ₂₃ , X ₂₄ , X ₂₅ , X ₃₁ , X ₃₂ , X ₃₃ , X ₃₄ , X ₃₅ , X ₄₁ , X ₄₂ , X ₄₃ , X ₄₄ , X _It consists of 25 pixels, which are ₄₅ , X ₅₁ , X ₅₂ , X ₅₃ , X ₅₄ , and X ₅₅ .

The predictive value calculation block obtained in the block generation unit 142 for predictive value calculation is supplied to the prediction unit 146.

Similar to the case of the class classification block, the number and shape of the pixels according to the predictive value calculation block are not limited to the examples described above. Note that the total number of pixels constituting the predictive value calculation block of the local decoder 127 is preferably configured to be larger than the total number of pixels constituting the class classification block.

Also, when the above-described block generation processing is performed, corresponding pixels do not exist near the image frame of the image. In the above case, for example, in the case where the processing operation is performed, assuming that the same pixels are performed (similar to the processing operation rather than the block generation), there are no corresponding pixels near the image frame of the image. In the above case, for example, suppose that the same pixels as the pixels constituting the image frame are located outside the image frame when the processing operation is performed.

The class classification unit 144 classifies the blocks supplied from the block generation unit 141 for class classification, and supplies these classified results to the prediction coefficient ROM 145 as class information. The prediction coefficient ROM 145 stores the prediction coefficient for each class therein, and when class information is received from the class classification unit 144, the prediction coefficient stored at the address corresponding to this class information is read out and predicted. Supplied to the unit 146. The prediction coefficients stored in the prediction coefficient ROM 145 are obtained by the method of the learning processing operation (prediction coefficient learning, which will be discussed later).

The prediction unit 146 calculates the prediction value of the raw image data (learning image data) and also the prediction coefficient read out from the prediction coefficient ROM 145 using the prediction value calculation block derived from the block generation unit 142 for calculating the prediction value. (Forecast)

Referring to the flowchart of FIG. 24, the operation of the local decoder 127 of FIG. 23 will be further described.

First, in the local decoder 127, in step S91, correction data derived from the level control unit 126 is sequentially received to be processed to generate blocks. In other words, in the block generation unit 141 for class classification, the correction data will be subdivided into a class classification block composed of five pixels at the center of the displayed correction data, and supplied to the class classification unit 144. In the block generation unit 142 for calculating the predicted value, the correction data will be subdivided into a predicted value calculating block composed of 5x5 pixels at the center of the displayed corrected data and supplied to the predicting unit 146.

It is also noted that in the block generation unit 141 for class classification and the block generation unit 142 for prediction value calculation, a corresponding prediction value calculation block and a corresponding class classification block are generated. This is because, in the block generation unit 141 for class classification, for example, when a class classification block composed of five pixels in the center around correction data X _{33 in} FIG. 7 is generated, the center around the same correction data X ₃₃ is generated. A predictive value calculation block consisting of 5x5 pixels at is generated.

In step S92, the class classification unit 144 executes a class classification processing operation on the class classification block, that is, detects a distribution state of respective levels of correction data in the block, so that the class classification block belongs. It is determined according to the class. This class determination result is supplied to the prediction coefficient ROM 145 as class information.

In the above embodiment of Fig. 24, the class classification processing operation is performed on a class classification block composed of five pixels in which one pixel is represented by two bits, so that each class classification block is any one of 512 classes (= (2 ² ) ⁵ ). It will be understood that they are classified as one.

The processing operation then proceeds to step S93, where the prediction coefficients are read from the address of the prediction coefficient ROM 145 corresponding to the class information supplied from the class classification unit 144. In step S94, the prediction unit 146 uses the 25 pixel values constituting the prediction value calculation block supplied from the prediction value calculation block according to the prediction coefficient and the following linear equation, to predict the values of "y" of the raw image data. Calculate the predicted values of E (y).

E (y) = w ₁ x ₁ + w ₂ x ₂ + ... w ₂₅ x ₂₅ + w _{0 '}

Where the symbols w ₁ , w ₂ , ..., w ₀ represent prediction coefficients, and the symbols x ₁ , x ₂ , ... represent pixel values (correction data) of the pixels constituting the predictive value calculation block. In particular, the symbol w _{0 '} is an offset coefficient added to represent E (y) in 8 bits because the raw 8-bit data "x _i " has been converted to 2-bit data.

In the embodiment of FIG. 23, as described above, the predicted values of 9 pixels may be calculated from 25 pixels constituting the predicted value calculation block.

More specifically, for example, class information C is outputted from the classification unit 144, the classification information C is 3x3 correction data _{X 22, X 23, X 24} , X 32, X 33, Associated with a class classification block consisting of X ₃₄ , X ₄₂ , X ₄₃ , and X ₄₄ , where the correction data X ₃₃ shown in FIG. 7 is centered. In addition, as the predictive value calculating block, the predictive value calculating block generating unit 142 is assumed. These predictive calculation blocks are X _{11 '} , X _12' , X _{13 '} , X _14' , X _{15 '} , X _21' , X _{22 '} , X _23' , X _{24 '} , X _25' , X _{31 '} , X _{32 '} , X _33' , X _{34 '} , X _35' , X _{41 '} , X _42' , X _{43 '} , X _44' , X _{45 '} , X _51' , X _{52 '} , X _53' , X _{54 '} , X _{55 '} , where correction data X ₃₃ is centered.

Moreover, it is assumed that w ₁ (k)-w ₂₅ (k), and w ₀ (k) are stored as a set of prediction coefficients at an address corresponding to class information C in the prediction coefficient ROM 145.

In the above case, the predicted values of the pixel values Y ₃₃ (1)-Y ₃₃ (9) of the 3x3 pixel (i.e., the part surrounded by the rectangle in FIG. 7) of the raw image data centered around the correction data X ₃₃ . E [Y ₃₃ (1)]-E [Y ₃₃ (9)] is calculated according to the following equation.

E [Y ₃₃ (k)] = w ₁ (k) X ₁₁ + w ₂ (k) X _{12 +} w ₃ (k) X ₁₃

+ w ₄ (k) X ₁₄ + w ₅ k) X ₁₅ + w ₆ (k) X ₂₁

+ w ₇ (k) X ₂₂ + w ₈ (k) X ₂₃ + w ₉ (k) X ₂₄

+ w ₁₀ (k) X ₂₅ + w ₁₁ (k) X ₃₁

+ w ₁₂ (k) X ₃₂ + w ₁₃ (k) X ₃₃

+ w ₁₄ (k) X ₃₄ + w ₁₅ (k) X ₃₅

+ w ₁₆ (k) X ₄₁ + w ₁₇ (k) X ₄₂

+ w ₁₈ (k) X ₄₃ + w ₁₉ (k) X ₄₄

+ w ₂₀ (k) X ₄₅ + w ₂₁ (k) X ₅₁

+ w ₂₂ (k) X ₅₂ + w ₂₃ (k) X ₅₃

+ w ₂₄ (k) X ₅₄ + w ₂₅ (k) X ₅₅ + w ₀ (k)

Nine of the prediction values are obtained in step S94 in the above-described method, and the processing operation returns to the previous step S91, so that the processing operations defined in step S94 in step S91 are repeated, The predictive values are calculated in units of 9 pixels.

Since the image processing apparatus for executing the learning operation (predictive coefficient learning) used to obtain the predictive coefficients stored in the predictive coefficient ROM 145 of FIG. 23 is configured in a similar method to that of FIG. 17, description thereof will be omitted.

Next, FIG. 27 shows an embodiment of the receiving device 4 "corresponding to the transmitting device 1 of FIG.

In the reception / reproducing apparatus 191, coded data recorded on the recording medium 2 is reproduced, or coded data transmitted via the transmission path 3 is received and supplied to the decoder 192.

Decoder 192 includes from block generation unit for class classification to prediction unit 198, which are prediction unit 146 in classification block generation unit 141 used for local decoder 127 shown in FIG. Corresponds to until. As a result, the prediction values at the decoder 192 are calculated from the correction data, and then the picture composed of these prediction values is output as a decoded picture in a manner similar to that of the local decoder 127 of FIG. The prediction coefficients for each class stored in the prediction coefficient ROM 197 are the same as the prediction coefficients used in the mapping coefficients for the learning process.

Since the correction data has error information less than the predetermined threshold value, it is possible to obtain an image which is actually the same as the original image data in the receiving device 4. In this embodiment, even though each process is performed for one frame, each process may be performed for one field, two frames, or two or more frames.

It should be understood that when normal interpolation is performed by an apparatus for decoding a thinned image by interpolation on the receiving side, the decoded image can be obtained even if the receiving apparatus 4 shown in Fig. 27 is not used. However, in this alternative case, the picture quality (resolution) of the final decoded picture is degraded.

Although the image for one frame has been divided to make the image block of the present embodiment, this block can be formed, for example, by collecting pixels placed at the same position in a plurality of frames which are continued in chronological order.

In this embodiment, the sum of squares of errors is used as the error information. Alternatively, for example, the sum of the absolute value of the error and the sum of the cube errors can be used as the error information. In consideration of the convergence characteristic, it is determined that the value is used as error information.

Fifth Embodiment

A fifth embodiment of the present invention is described with reference to the transmission device of FIG. 18, the learning device for the mapping coefficients of FIG. 25, and the local decoder of FIG. 23. The receiving apparatus used in the fifth embodiment is explained in FIG. Unless otherwise noted, the transmitting device of FIG. 18, the local decoder of FIG. 23, and the receiving device of FIG. 27 operate as previously described for the fourth embodiment of the present invention.

Referring to FIG. 25, a learning apparatus for mapping coefficients used in the fifth embodiment is described. FIG. 25 shows the structure of an image processing apparatus that performs learning processing (mapping coefficient learning) used to calculate mapping coefficients for storage in the mapping coefficient memory 114 of FIG.

Note that, for example, according to the image processing apparatus of FIG. 21, even when the function "f" is represented by a linear equation and further represented by a quadratic or higher equation, an optimal prediction coefficient can be obtained. have. On the contrary, in the image processing apparatus of FIG. 25, the optimum prediction coefficient can be calculated only when the function "f" is represented by a linear equation.

In other words, the image processing apparatus of FIG. 25 can be used in the following cases. That is, in FIG. 18, the pixel values of respective pixels constituting the block made of 3 × 3 pixels output from the block generation unit 111 are set to y ₁ , y ₂ ----, y ₉ , and mapping. When the mapping coefficient output from the coefficient memory 114 is set to k ₁ , k ₂ ---, k ₉ , the calculation unit 116 obtains the function "f" (according to the format mentioned below to obtain correction data). Calculate y ₁ , y ₂ , ---, k ₁ , k ₂ ----).

f (˙) = k ₁ y ₁ + k ₂ y ₂ + ... · k ₉ y ₉

The optimal correction data calculation unit 180 is supplied with image data suitable for the learning operation, for example, in units of one frame. The optimal correction data calculation unit 160 includes a compression unit 170, a level limiting unit 171, and a correction unit 172. A local decoder 173, an error calculation unit 174, and a determination unit 175. The optimal correction data calculation unit 160 is a value for constructing an optimal image in which the number of pixels used to predict the original image data from the learning image fed into the optimal correction data calculation unit 160 is reduced (if necessary, &Quot; optimum correction data " is calculated to supply the calculated pixel value to the buffer unit 176.

In other words, the training image data supplied to the optimum correction data calculation unit 160 is supplied to the compression unit 170 and the error calculation unit 174. The compression unit 170 simply reduces the training image data at the same speed as the pixel is thinned by the calculation unit 116 in Fig. 18, so that the training image data is compressed and the compressed image data is level limited. Supplied to the unit 171. In other words, in the present embodiment, the learning image data is simply reduced by 1/9 (when nine pixels of 3x3 pixels are set to one block, only the center pixel of the block is extracted). In the level limiting unit 171, after the level of the pixel data is limited, the level limiting pixel data is output to the correction unit 172. The correction unit 172 corrects the data under the control of the determination unit 175, so that the data is supplied from the level limiting unit 171, and simply reduced and compressed by selecting 2 bits on the MSB side from 8-bit data. The level of data (called "compressed data", if necessary) is limited to 2 bits. Then, the corrected final data of the correction unit 172 is supplied to the local decoder 173 so that the corrected data, hereinafter referred to as " correction data ", if necessary, is the center pixel in the block composed of 3 x 3 pixels. The pixel value of is corrected in the same manner as output from the calculation unit 116 of FIG.

Similar to the case of the error calculator 138 of FIG. 21, the error calculating unit 174 calculates a prediction error of prediction values derived from the local decoder 173 with respect to the original image data, which predicts the error. It is put into the unit. This prediction error is supplied to the determination unit 175 as error information.

Based on the error information derived from the error calculation unit 174, the determination unit 175 makes a judgment as to whether or not the correction data output from the correction unit 172 is the same as the original image data compressed properly. Then, when the determination unit 175 determines that the correction data output from the correction unit 172 is not appropriately the same as the compression result of the original image data, the determination unit 175 controls the correction unit 172 and Further, the compressed data is corrected, and the newly obtained correction data is output. In addition, when the determination unit 175 makes a judgment as to whether or not the correction data output from the correction unit 172 is appropriately equal to the compression result of the original image data, the correction data supplied from the correction unit 172 is The buffer unit 176 is supplied as optimum correction data.

The buffer unit 176 includes a memory 176A therein for storing optimum correction data supplied from the optimum correction data calculation unit 160. Moreover, the buffer unit 176 reads out the optimum correction data corresponding to the center pixel of the block read out from the memory 177A of the block generation unit 177, among the optimum correction data stored in the memory 176A. Note that when correction data for one frame is stored in the memory 176A, the buffer unit 176 outputs a control signal indicating the data storage to the block generation unit 177.

Similar to the optimum correction data calculation unit 160, the training block data is supplied to the block generation unit 177 in units of one frame. The block generation unit 177 includes a memory 177A for storing learning image data provided therein. Upon receiving the control signal transmitted from the buffer unit 176, the block generation unit 177 configures the learning image data stored in the memory 177A in the same manner as the block generation unit 111 of FIG. Repartition into blocks. The blocks are then read sequentially and supplied to the ADRC processing unit 178 and the memory 180.

When a block is read from the memory 177A included in the memory unit 177, the block generating unit 177 supplies a control signal indicating the position of the block to the buffer unit 176. In the buffer unit 176, a block of 3x3 pixels read out from the memory 177A is recognized in response to the control signal, and as previously described, the optimal correction data corresponding to the center pixel of the block is stored in the memory ( 176A). That is, as a result, the specific block composed of 3x3 pixels corresponding to the block and the optimum correction data are simultaneously supplied to the memory 180.

The ADRC processing unit 178 and the class classification unit 179 are disposed similarly to the ADRC processing unit 112 and the class classification unit 113 of FIG. 18, respectively. The class information for the block derived from the block generation unit 177, output from the class classification unit 179, is supplied to the memory 180 as an address.

The memory 180 has a corresponding relationship at the address corresponding to the optimum correction data supplied from the buffer unit 176 and the class information supplied from the class classification unit 179 therein, from the block generation unit 177. The supplied block is stored therein. Note that in the memory 180, a plurality of pieces of information may be stored at one address. As a result, a plurality of sets of both optimal correction data and blocks corresponding to specific class information can be stored.

The calculation unit 181 quantums nine pixels y ₁ , y ₂ ,..., Y ₉ composed of 3 × 3 pixels of the learning image data stored in the memory 180 and the optimal correction data y ′ corresponding to the block. Read it. Then, the least squares method is applied so that all classes of mapping coefficients k ₁ to k ₉ are calculated and supplied to memory 182. The memory 182 stores therein all classes of mapping coefficients k ₁ to k ₉ supplied from the calculation unit 181 at an address corresponding to the class.

Referring to the flowchart shown in FIG. 26, the operation of the image processing apparatus of FIG. 25 is further described.

When the training image data is input, the training image data is stored in the memory 177A of the block generation unit 177 and supplied to the optimum correction data calculation unit 160. In step S101, when the optimum correction data calculation unit 160 receives the learning image data, the optimum correction data calculation unit 160 calculates the optimum correction data for the learning image data.

The processing operation defined in step S101 is similar to the processing operation of the flowchart shown in FIG. That is, first, in step S1, the compression unit 170 generates compressed data by reducing the learning curve to 1/9. In step S2, the data is limited by the level limiting unit 171, after which the level limiting data is first outputted to the local decoder 173 through the correction unit 172 without any correction. In the local decoder 173, in step S3, correction data supplied from the correction unit 172 (i.e., the local decoding operation is performed) (as described above, is simply generated by initially thinning the image data). The estimated value of the original image data is calculated based on the compressed data). These predictions are supplied to the error calculator 174.

If the predictions of the original picture data are received from the local decoder 173, the error calculator 174 calculates the error of the predictions of the prediction values slaughtered from the local decoder 173 with respect to the original picture data in step S4 and error. This side error calculated as the information is supplied to the determination unit 175. Upon receiving the error information from the error calculator 174, the determination unit 175 causes the correction data output from the correction unit 172 to suit the compression result of the original image data based on the error information in step S5. Make a judgment about whether it is the same.

In other words, in step S5, a check is made as to whether or not the error information is smaller than the predetermined threshold value "?". If it is determined in step S5 that the error information is smaller than the predetermined threshold "ε", it is recognized that the correction data output by the correction unit 172 is suitably the same as the compression result of the raw image data. The processing operation then shifts to step S6 in which the determination unit 175 controls the correction unit 172 to correct the compressed data output from the level control unit 171. The correction unit 172 changes the correction amount (correction value "Δ") to correct the compressed data under the control of the determination unit 175. The final correction data is output to the local decoder 173. The processing operation then returns to step S3 in which a similar processing operation is repeated.

Note that the compressed data can be corrected in the same way, for example, by changing the mapping coefficient as previously described in FIG. 21.

On the other hand, if it is determined in step S5 that the error information is smaller than the predetermined threshold "ε", the determination unit 175 determines that the correction data output by the correction unit 172 is appropriately the original image. Recognizing that the data is the same as the correction result, the correction unit 172 causes the correction unit 172 to output the correction data generated when the error information is smaller than the predetermined threshold value "?" To the buffer unit 176 as optimum correction data. do. Then, the optimum correction data is stored in the memory 176A included in the buffer unit 176, and the processing operation is returned.

As described previously, the correction data obtained by correcting the compressed data obtained when the error information is smaller than the predetermined threshold value "?" Can be stored in the memory 176A as optimum correction data. It should be noted that since the optimal correction data makes the error information smaller than the predetermined threshold "ε", it is possible to produce an image that is actually identical to the raw image data when the prediction value is calculated by using the optimal correction data. .

Referring again to FIG. 26, when the buffer unit 176 stores the optimal correction data for one frame in the memory 176A, the buffer unit 176 outputs a control signal to the block generation unit 177. . In step S102, upon receiving the control signal derived from the buffer unit 176, the block generation unit 177 reads a block of learning image data stored in the memory 177A, and converts the read block into an ADRC processing unit. 178 and the memory 180.

At the same time, when a block is read from the memory 177A, the block generating unit 177 supplies a control signal indicating the position of the block to the buffer unit 176. In response to the control signal, the buffer unit 176 recognizes a block of 3x3 pixels read from the memory 177A, reads the optimum correction data corresponding to the center pixel of the block, and reads the optimum correction. The data is supplied to the memory 180.

Then, the block derived from the block generation unit 177 is subjected to ADRC processing, and further, the process shifts to step S103 in which the block is class classified in the class classification unit 179. The classification result is supplied to the memory 180 as an address.

In step S104, the memory 180 has a corresponding relationship at an address corresponding to optimal correction data provided to the buffer unit 176 and class information supplied from the class classification unit 179 therein. The block (learning data) supplied from the block generating unit 177 is stored.

The processing operation then shifts to step S105 where a check is made as to whether a block for one frame and optimal correction data are stored in the memory 180. In step S105, if it is determined that the block for one frame and the optimal correction data have not yet been stored in the memory 180, the next block is read from the block generating unit 177, and also corresponds to the read block. The optimal correction data is read from the buffer unit 176. The processing operation then returns to step S103, which is defined after step S103 is repeated.

Conversely, in step S105, when it is determined that the block for one frame and the optimum correction data are stored in the memory 180, the processing operation is checked whether or not the processing operations for all learning images are completed. Transferred to step S106 to be performed. If it is determined in step S106 that all learning image data has not yet been terminated, the processing operation is shifted to step S101, in which processing operations defined after step S101 are repeated for the next learning image. do.

On the other hand, if it is determined in step S106 that the processing operations for all the learning image data have been completed, the processing operation reads the optimal correction data and the block stored in the memory 180 by the calculation unit 181 for every class. Thus, the general equation shown in format (7) based on the data and blocks is set. Moreover, in step S108, the calculation unit 181 solves this general equation to calculate mapping coefficients that can minimize the error for every class. In step S109, the drawing number is supplied to the memory 182 and stored therein, and then the processing operation is terminated.

When the function " f " is represented by a linear equation, the mapping coefficients stored in the memory 182 in the manner described above are stored in the mapping coefficient memory 114 of Fig. 18, and the optimal correction data uses the mapping coefficients. Can be obtained.

Note that there are the following difficulties. That is, a large number of general equations that could obtain mapping coefficients could not be obtained. In this case, in the calculating unit 116 of Fig. 18, such a mapping coefficient is set to an omission value, that is, k ₁ to k ₂ = 1/9. For example, an average value of nine pixels constituting a block of 3x3 pixels output from the block generation unit 111 is output.

Although the present invention has been described with respect to an image processing apparatus, such an image processing apparatus may have particular advantages in the following cases, for example, when a standardized television signal (NTSC TV signal) is coded, and a large amount of So-called "high precision" television signals containing data are decoded.

In summary, the configuration of the first embodiment of the present invention includes the transmission device of FIG. 3, the compression unit of FIG. 5, the local decoder of FIG. 8, the error calculator of FIG. 11, and the determination unit of FIG. 13. The receiving device of Fig. 15 is used in the first embodiment. The second embodiment of the present invention includes a transmission device of FIG. 28, a compression unit of FIG. 5, a local decoder of FIG. 16, an error calculator of FIG. 11, a determination unit of FIG. 13, and a learning device for prediction coefficients of FIG. 17. do. The receiving device of FIG. 15 is used in the second embodiment. The configuration of the third embodiment of the present invention includes the transmission device of FIG. 30, the compression unit of FIG. 5, the local decoder of FIG. 32, the error calculator of FIG. 11, and the determination unit of FIG. 34. The receiving device of Fig. 36 is used in the third embodiment. The configuration of the fourth embodiment of the present invention includes a learning apparatus for mapping coefficients of FIG. 21, a local decoder of FIG. 23, and a learning apparatus for prediction coefficients of FIG. 17. The receiving device of FIG. 27 is used in the fifth embodiment.

Although only a few embodiments have been described in detail, those skilled in the art will clearly understand that various modifications are possible without departing from the technology. All such changes will fall within the scope of the following claims.

According to the present invention, there is an effect that a highly efficient, high quality predictive image can be generated by performing less data processing operation.

1 is a block diagram showing a configuration of an embodiment of the present invention applied to an image processing system.

2 is a block diagram showing an exemplary configuration of a transmission device 1.

3 is a block diagram of a first embodiment of the transmission device 1 of FIG. 2 (excluding the transmission / recording device 16).

4 is a flowchart illustrating the operation of the transmission device of FIG.

5 is a block diagram of a compression unit 21 of each of FIGS. 3, 19 and 21.

6 is a flowchart illustrating the operation of the compression unit 21 of FIG. 5.

7 is a process diagram of the thinning unit 31 of FIG. 5.

8 is a block diagram of a local decoder 22 of FIG.

9 (a) and 9 (b) are class classification processing diagrams.

10 is a flowchart illustrating operation of the local decoder 22 of FIG.

11 is a block diagram of the error calculator 23 of FIGS. 3 and 28.

12 is a flowchart illustrating the operation of the error calculator 23 of FIG.

FIG. 13 is a block diagram of the determination unit 24 of FIG. 3.

14 is a flowchart illustrating the operation of the determination unit of FIG.

15 is a block diagram of the receiving device of FIG. 1 used in connection with the transmitting device of FIG.

FIG. 16 is a block diagram of the local decoder 22 'of FIG.

FIG. 17 is a block diagram of an embodiment of an image processing apparatus that calculates prediction coefficients stored in prediction coefficient ROM 81 of FIG.

18 is a block diagram of the transmission device 1 of FIG. 2 (except for the transmission / recording device 16) according to the fourth and fifth embodiments of the present invention.

19 (a), 19 (b) and 19 (c) are ADRC processing diagrams.

20 is a flowchart illustrating the operation of the transmission device of FIG. 18;

21 is a block diagram of an image processing apparatus according to a fourth embodiment, for performing a learning operation used to obtain mapping coefficients.

22 is a flowchart illustrating the operation of the image processing apparatus of FIG. 21;

FIG. 23 is a block diagram of the local decoder 127 of FIG.

24 is a flowchart illustrating operation of the local decoder 127 of FIG.

25 is a block diagram of an image processing apparatus according to a fifth embodiment, performing a learning operation used to obtain mapping coefficients.

FIG. 26 is a flowchart illustrating the operation of the image processing apparatus of FIG. 25;

27 is a block diagram of the receiving device of FIG. 1 used in connection with the transmitting device of FIG.

28 is a block diagram of a transmission device according to a second embodiment of the present invention.

29 is a flowchart illustrating the operation of the transmission device of FIG. 28;

30 is a block diagram of a transmission apparatus according to a third embodiment of the present invention.

FIG. 31 is a flowchart illustrating an operation of the transmitting apparatus of FIG. 30;

32 is a block diagram of a local decoder 22 '' of the transmitting apparatus of FIG.

33 is a flowchart illustrating operation of the local decoder 22 '' in FIG.

34 is a block diagram of a determination device 24 'of the transmission device of FIG.

35 is a flowchart illustrating the operation of the transmission device of FIG. 34;

36 is a block diagram of the receiving device of FIG. 1 used in connection with the transmitting device of FIG.

<Explanation of symbols for the main parts of the drawings>

1: transmission device

2: recording medium

3: transmission unit

4: receiving device

11: I / F

12: ROM

13: RAM

14: CPU

15: external memory

16: transmission / recording device

21: compression unit

22: local decoder

23: Error Calculator

24: judgment unit

25: multiplexer

30 thinning unit

31: level limit unit

32: correction unit

33: Correction value ROM

41: block generation unit for class classification

42: block generation unit for prediction value calculation

43: Class Classification Adaptive Processing Circuit

45: class classification unit

46: adaptive processing unit

51: block generation unit

52: squared error calculation unit

53, 54: Calculator

55: integrator

56: memory

61: prediction coefficient memory

62: correction data memory

63: error information memory

64: comparator

65: controller

71: reception / playback unit

72: separation unit

73: block generation unit for class classification

75: class classification unit

76: prediction unit

77: block generation unit for prediction value calculation

81: prediction coefficient ROM

82: prediction unit

91: learning block generation unit

92: teacher block generation unit

94: class classification unit

95: switch

96: learning data memory

97: counter

98: teacher data memory

99: calculation unit

100: memory

Claims (104)

An apparatus for coding image data having a plurality of pixels, Means for reducing the number of pixels of raw image data to produce compressed data; Means for generating limited compressed data by limiting levels of pixels of the compressed data; Means (correction means) for correcting the limited compressed data to produce corrected data; Means (prediction means) for predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; Means (calculating means) for calculating a prediction error of the prediction data with respect to the raw image data; Means (judging means) for determining suitability of the correction data as coded data based on the prediction error; And Means for outputting the correction data as coded data of the raw image data Including; And the correcting means, the predicting means, the calculating means and the determining means are repeatedly operated based on the determined suitability. The method of claim 1, wherein the prediction means Means for generating class information of the correction data; And Means for generating the prediction pixels according to the class information Image data coding apparatus comprising a. The method of claim 1, wherein the prediction means Means for generating prediction coefficients based on the correction data; And Means for generating the prediction pixels based on the correction data and the prediction coefficients Image data coding apparatus comprising a. The method of claim 1, wherein the prediction means Means for generating class information based on the plurality of pixels of the correction data; Means for generating prediction coefficients for each class based on the raw image data and the correction data; And Means for generating the plurality of prediction pixels based on the prediction coefficients and the correction data corresponding to respective class information. Image data coding apparatus comprising a. 5. An image data coding apparatus according to claim 4, wherein said output means outputs said coding data together with said prediction coefficients for each class. The method of claim 1, wherein the prediction means A memory for storing prediction coefficients for each class; Means for generating class information based on the plurality of pixels of the correction data; And Means for reading the prediction coefficients corresponding to each of the class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data. Image data coding apparatus comprising a. 7. The image data coding apparatus according to claim 6, wherein the prediction coefficients for each class stored in the memory are generated based on learning raw image data. 8. An image data coding apparatus according to claim 7, wherein said output means outputs said coding data together with said prediction coefficients for each class. An image data coding apparatus according to claim 1, wherein said correction means includes a memory for storing correction values and corrects said compressed data using said correction values. 2. The apparatus of claim 1, wherein the determining means determines the suitability of the correction data by detecting whether the prediction error is smaller than a predetermined threshold value; And the output means outputs the correction data as the coding data when the prediction error is smaller than the predetermined threshold value. An apparatus for decoding coded data generated by coding image data, the apparatus comprising: Reducing the number of pixels of the raw image data to generate compressed data, Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step), Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step), Calculating a prediction error of the prediction data with respect to the raw image data (calculation step), Determining the suitability of the correction data as coded data based on the prediction error (judgment step), Based on the determined suitability, repeating the correction step, the prediction step, the calculation step and the determination step, and Outputting the correction data as coding data Means for receiving the coded data generated by; And Means for decoding the coded data to produce decoded picture data Image data decoding apparatus comprising a. 12. The apparatus of claim 11, wherein said decoding means Means for generating class information of the coding data; And Means for generating the decoded picture data according to the class information Image data decoding apparatus comprising a. 12. The apparatus of claim 11, wherein the coding data includes prediction coefficients for generating the decoded picture data, And said decoding means comprises means for generating said decoded picture data based on said prediction coefficients and said coding data. 12. The apparatus of claim 11, wherein the coding data includes prediction coefficients for each class to produce the decoded picture data, and the decoding means Means for generating class information using a plurality of pixels of the coding data; And Means for generating the decoded picture data using the prediction coefficients corresponding to the coding data. Image data decoding apparatus comprising a. 12. The apparatus of claim 11, wherein said decoding means A memory for storing prediction coefficients for each class; Means for generating class information using a plurality of pixels of the coding data; And Means for reading the prediction coefficients corresponding to the class information from the memory and generating the prediction pixel using the read prediction coefficients and the coding data Image data decoding apparatus comprising a. 16. The image data decoding apparatus of claim 15, wherein the prediction coefficients for each class stored in the memory are generated using training raw image data. In the method for coding image data having a plurality of pixels, Reducing the number of pixels of the raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step); Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step); Calculating a prediction error of the prediction data with respect to the raw image data (calculation step); Determining the suitability of the correction data as coded data based on the prediction error (judgment step); Repeating the correction step, the prediction step, the calculation step and the determination step based on the determined suitability; And Outputting the correction data as coding data of the raw image data Image data coding method comprising a. 18. The method of claim 17, wherein the predicting step Generating class information of the correction data; And Generating the prediction pixels according to the class information Image data coding method comprising a. 18. The method of claim 17, wherein the predicting step Generating prediction coefficients based on the correction data; And Generating the prediction pixels based on the correction data and the prediction coefficients Image data coding method comprising a. 18. The method of claim 17, wherein the predicting step Generating class information based on the plurality of pixels of the correction data; Generating prediction coefficients for each class based on the raw image data and the correction data; And Generating the plurality of prediction pixels based on the prediction coefficients and the correction data corresponding to each class information Image data coding method comprising a. 21. The method of claim 20, further comprising outputting the coding data with prediction coefficients for each class. 18. The method of claim 17, wherein the predicting step Storing prediction coefficients for each class; Generating class information based on the plurality of pixels of the correction data; And Reading prediction coefficients corresponding to respective class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data Image data coding method comprising a. 23. The method of claim 22, wherein the stored prediction coefficients for each class are generated based on the learning raw image data. 24. The method of claim 23, wherein the coding data is output together with the prediction coefficients for each class. 18. The method of claim 17, wherein the correcting step includes storing correction values, wherein the compressed data is corrected using the correction values. 18. The method of claim 17, wherein the suitability determination step is Detecting whether the prediction error is less than a predetermined threshold; And Outputting the correction data as the coding data when the prediction error is smaller than the predetermined threshold value Image data coding method comprising a. A method for decoding coded data generated by coding picture data, the method comprising: Reducing the number of pixels of the raw image data to generate compressed data, Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step), Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step), Calculating a prediction error of the prediction data with respect to the raw image data (calculation step), Determining the suitability of the correction data as coded data based on the prediction error (judgment step), Based on the determined suitability, repeating the correction step, the prediction step, the calculation step and the determination step, and Outputting the correction data as coding data of the raw image data Receiving the coded data generated by; And Decoding the coded data to generate decoded picture data Image data decoding method comprising a. 28. The method of claim 27, wherein decoding the coded data Generating class information of the coding data; And Generating the decoded picture data according to the class information. Image data decoding method comprising a. 28. The apparatus of claim 27, wherein the coded data comprises prediction coefficients for generating the decoded picture data, and wherein the step of decoding the coded data is based on the prediction coefficients and the coded data. And generating the data. 28. The method of claim 27, wherein the coded data includes prediction coefficients for each class to produce the decoded picture data, and wherein decoding the coded data comprises: Generating class information using a plurality of pixels of the coding data; And Generating the decoded picture data using the prediction coefficients corresponding to the coded data. Image data decoding method comprising a. 28. The method of claim 27, wherein decoding the coded data Storing prediction coefficients for each class; Generating class information using a plurality of pixels of the coding data; And Reading the prediction coefficients corresponding to the class information from the memory and generating the prediction pixel using the read prediction coefficients and the coding data Image data decoding method comprising a. 32. The method of claim 31, wherein the stored prediction coefficients for each class are generated using learning raw image data. An apparatus for coding image data having a plurality of pixels, A subsampling circuit for reducing the number of pixels of raw image data to generate compressed data; Limiting circuitry for generating limited compressed data by limiting levels of pixels of the compressed data; A correction circuit for correcting the limited compressed data to produce corrected data; A prediction circuit for predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels; A calculator for calculating a prediction error of the prediction data with respect to the raw image data; Determination circuitry for determining suitability of the correction data as coded data based on the prediction error; And An output circuit for outputting the correction data as coded data of the raw image data Including; And the correction circuit, the prediction circuit, the calculator, and the determination circuit are repeatedly operated based on the determined suitability. 34. The method of claim 33 wherein the prediction circuit is Circuitry for generating class information of the correction data; And Circuitry for generating the prediction pixels according to the class information Image data coding apparatus comprising a. 34. The method of claim 33 wherein the prediction circuit is Circuitry for generating prediction coefficients based on the correction data; And Circuitry for generating the prediction pixels based on the correction data and the prediction coefficients Image data coding apparatus comprising a. 34. The method of claim 33 wherein the prediction circuit is Circuitry for generating class information based on the plurality of pixels of the correction data; Circuitry for generating prediction coefficients for each class based on the raw image data and the correction data; And Circuitry for generating the plurality of prediction pixels based on the prediction coefficients and the correction data corresponding to each class information Image data coding apparatus comprising a. The image data coding apparatus according to claim 36, wherein the output circuit outputs the coding data together with the prediction coefficients for each class. 34. The method of claim 33 wherein the prediction circuit is A memory for storing prediction coefficients for each class; Circuitry for generating class information based on the plurality of pixels of the correction data; And Circuitry for reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data Image data coding apparatus comprising a. The image data coding apparatus according to claim 38, wherein the prediction coefficients for each class stored in the memory are generated based on learning raw image data. 40. The image data coding apparatus according to claim 39, wherein the output circuit outputs the coding data together with the prediction coefficients for each class. 34. The image data coding apparatus of claim 33, wherein the correction circuit includes a memory for storing correction values and corrects the compressed data using the correction values. 34. The apparatus of claim 33, wherein the determination circuit determines the suitability of the correction data by detecting whether the prediction error is smaller than a predetermined threshold value, And the output circuit outputs the correction data as the coding data when the prediction error is smaller than the predetermined threshold. A method for transferring image data having a plurality of pixels, Reducing the number of pixels of the raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step); Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step); Calculating a prediction error of the prediction data with respect to the raw image data (calculation step); Determining the suitability of the correction data as coded data based on the prediction error (judgment step); Repeating the correction step, the prediction step, the calculation step and the determination step based on the determined suitability; Outputting the correction data as coding data of the raw image data; And Transmitting the output correction data. Image data transmission method comprising a. 44. The method of claim 43, wherein the predicting step Generating class information of the correction data; And Generating the prediction pixels according to the class information Image data transmission method comprising a. 44. The method of claim 43, wherein the predicting step Generating prediction coefficients based on the correction data; And Generating the prediction pixels based on the correction data and the prediction coefficients Image data transmission method comprising a. 44. The method of claim 43, wherein the predicting step Generating class information based on the plurality of pixels of the correction data; Generating prediction coefficients for each class based on the raw image data and the correction data; And Generating the plurality of prediction pixels based on the prediction coefficients and the correction data corresponding to each class information Image data transmission method comprising a. 47. The method of claim 46, further comprising outputting the coding data with the prediction coefficients for each class. 44. The method of claim 43, wherein the predicting step Storing prediction coefficients for each class; Generating class information based on the plurality of pixels of the correction data; And Reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data Image data transmission method comprising a. 49. The method of claim 48, wherein the stored prediction coefficients for each class are generated based on learning raw image data. 50. The method of claim 49, wherein the coding data is output together with the prediction coefficients for each class. 44. The method of claim 43, wherein the correcting step includes storing correction values, wherein the compressed data is corrected using the correction values. 44. The method of claim 43, wherein said determining conformance step Detecting whether the prediction error is less than a predetermined threshold; And Outputting the correction data as the coding data when the prediction error is smaller than the predetermined threshold value Image data transmission method comprising a. A recording medium on which coded digital signals are recorded, Reducing the number of pixels of the raw image data to generate compressed data; Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step); Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step); Calculating a prediction error of the prediction data with respect to the raw image data (calculation step); Determining the suitability of the correction data as coded data based on the prediction error (judgment step); Repeating the correction step, the prediction step, the calculation step and the determination step based on the determined suitability; Outputting the correction data as coding data of the raw image data; And Recording the coded data on the recording medium Recording medium prepared by. 54. The method of claim 53 wherein the predicting step Generating class information of the correction data; And Generating the prediction pixels according to the class information Recording medium comprising a. 54. The method of claim 53 wherein the predicting step Generating prediction coefficients based on the correction data; And Generating the prediction pixels based on the correction data and the prediction coefficients Recording medium comprising a. 54. The method of claim 53 wherein the predicting step Generating class information based on the plurality of pixels of the correction data; Generating prediction coefficients for each class based on the raw image data and the correction data; And Generating the plurality of prediction pixels based on the prediction coefficients corresponding to the information for each class and the correction data Recording medium comprising a. 54. The recording medium of claim 53, further comprising outputting the coding data with the prediction coefficients for each class. 54. The method of claim 53 wherein the predicting step Storing prediction coefficients for each class; Generating class information based on the plurality of pixels of the correction data; And Reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data Recording medium comprising a. 59. The recording medium of claim 58, wherein the stored prediction coefficients for each class are generated based on learning raw image data. 60. The recording medium of claim 59, wherein the coding data is output together with the prediction coefficients for each class. 54. The recording medium of claim 53, wherein the correcting step includes storing correction values, and the compressed data is corrected using the correction values. 54. The method of claim 53, wherein said determining conformance step Detecting whether the prediction error is less than a predetermined threshold; And Outputting the correction data as the coding data when the prediction error is smaller than the predetermined threshold Recording medium comprising a. In a method for coding data having a plurality of samples, Reducing the number of samples of raw data to produce compressed data; Limiting the level of samples of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step); Predicting the raw image data according to the correction data to generate predictive data having a plurality of predictive samples (prediction step); Calculating a prediction error of the prediction data with respect to the raw data (calculation step); Determining the suitability of the correction data as coding data on the basis of the prediction error (decision step); Repeating the correction step, the prediction step, the calculation step and the determination step based on the determined suitability; And Outputting the correction data as coding data of the raw data Data coding method comprising a. 66. The method of claim 63, wherein the predicting step Generating class information of the correction data; And Generating the prediction samples according to the class information Data coding method comprising a. 64. The method of claim 63 wherein the predicting step Generating prediction coefficients based on the correction data; And Generating the prediction samples based on the correction data and the prediction coefficients Data coding method comprising a. 66. The method of claim 63, wherein the predicting step Generating class information based on the plurality of samples of the correction data; Generating prediction coefficients for each class based on the raw data and the correction data; And Generating the plurality of prediction samples based on the prediction coefficients and the correction data corresponding to respective class information. Data coding method comprising a. 67. The method of claim 66, further comprising outputting the coding data with the prediction coefficients for each class. 66. The method of claim 63, wherein the predicting step Storing prediction coefficients for each class; Generating class information based on the plurality of samples of the correction data; And Reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction samples based on the read prediction coefficients and the correction data Data coding method comprising a. 69. The method of claim 68, wherein the stored prediction coefficients for each class are generated based on learning raw data. 70. The method of claim 69, wherein the coding data is output together with the prediction coefficients for each class. 64. The method of claim 63, wherein the correcting step includes storing correction values, wherein the compressed data is corrected using the correction values. 66. The method of claim 63, wherein the determining of conformance Detecting whether the prediction error is less than a predetermined threshold; And Outputting the correction data as the coding data when the prediction error is smaller than the predetermined threshold value Data coding method comprising a. A method for decoding coded data generated by coding data, the method comprising: Limiting the level of samples of compressed data to produce limited compressed data, Correcting the limited compressed data to generate corrected data (correction step), Predicting the raw image data according to the correction data to generate predictive data having a plurality of predictive samples (prediction step), Calculating a prediction error of the prediction data with respect to the raw data (calculation step), Determining the suitability of the correction data as coding data on the basis of the prediction error (decision step), Based on the determined suitability, repeating the correction step, the prediction step, the calculation step and the determination step, and Outputting the correction data as coding data of the raw data Receiving the coded data generated by; And Decoding the coded data to generate decoded data Data decoding method comprising a. 74. The method of claim 73, wherein decoding the coded data comprises Generating class information of the coding data; And Generating the decoded data according to the class information. Data decoding method comprising a. 74. The method of claim 73, wherein the coded data includes prediction coefficients for generating the decoded data, and wherein decoding the coded data generates the decoded data based on the prediction coefficients and the coded data. Data decoding method comprising a. 74. The method of claim 73, wherein the coded data includes prediction coefficients for each class to produce the decoded data, and wherein decoding the coded data comprises: Generating class information using a plurality of samples of the coding data; And Generating the decoded data using the prediction coefficients corresponding to the coded data. Data decoding method comprising a. 74. The method of claim 73, wherein decoding the coded data comprises Storing prediction coefficients for each class; Generating class information using a plurality of samples of the coding data; And Reading the prediction coefficients corresponding to the class information from the memory and generating the prediction sample using the read prediction coefficients and the coding data Data decoding method comprising a. 78. The method of claim 77, wherein the stored prediction coefficients for each class are generated using learning raw data. An apparatus for coding data having a plurality of samples, the apparatus comprising: A subsampling circuit for reducing the number of samples of raw data to produce compressed data; Limiting circuitry for limiting the level of samples of the compressed data to produce limited compressed data; A correction circuit for correcting the limited compressed data to generate correction data; A prediction circuit for predicting the raw data according to the correction data to generate prediction data having a plurality of prediction samples; A calculator for calculating a prediction error of the prediction data with respect to the raw data; Determination circuitry for determining suitability of the correction data as coded data based on the prediction error; And An output circuit for outputting the correction data as coded data of the raw data Including; And the correction circuit, the prediction circuit, the calculator and the determination circuit are repeatedly operated based on the determined suitability. 80. The apparatus of claim 79, wherein the prediction circuit is Circuitry for generating class information of the correction data; And Circuitry for generating the prediction samples according to the class information Data coding apparatus comprising a. 80. The apparatus of claim 79, wherein the prediction circuit is Circuitry for generating prediction coefficients based on the correction data; And Circuitry for generating the prediction samples based on the correction data and the prediction coefficients Data coding apparatus comprising a. 80. The apparatus of claim 79, wherein the prediction circuit is Circuitry for generating class information based on a plurality of samples of the correction data; Circuitry for generating prediction coefficients for each class based on the raw data and the correction data; And Circuitry for generating the plurality of prediction samples based on the prediction coefficients and the correction data corresponding to respective class information Data coding apparatus comprising a. 83. The data coding apparatus of claim 82, wherein the output circuit outputs the coding data along with the prediction coefficients for each class. 80. The apparatus of claim 79, wherein the prediction circuit is A memory for storing prediction coefficients for each class; Circuitry for generating class information based on a plurality of samples of the correction data; And Circuitry for reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction samples based on the read prediction coefficients and the correction data Data coding apparatus comprising a. 86. The data coding apparatus of claim 84, wherein the prediction coefficients for each class stored in the memory are generated based on learning raw data. 86. The data coding apparatus of claim 85, wherein the output circuit outputs the coding data along with the prediction coefficients for each class. 80. The data coding apparatus of claim 79, wherein the correction circuit includes a memory for storing correction values and corrects the compressed data using the correction values. 80. The apparatus of claim 79, wherein the determining circuit determines the suitability of the correction data by detecting whether the prediction error is smaller than a predetermined threshold value, And the output circuit outputs the correction data as the coding data when the prediction error is smaller than the predetermined threshold value. An apparatus for decoding coded data generated by coding data, the apparatus comprising: Reducing the number of samples of raw data to produce compressed data, Correcting the compressed data to generate correction data (correction step), Predicting the raw data according to the correction data to generate prediction data having a plurality of prediction samples (prediction step), Calculating a prediction error of the prediction data with respect to the raw data (calculation step), Determining the suitability of the correction data as coded data based on the prediction error (judgment step), Based on the determined suitability, repeating the correction step, the prediction step, the calculation step and the determination step, and Outputting the correction data as coding data of the raw data Circuitry for receiving the coded data generated by; And A decoder for decoding the coded data to produce decoded data Data decoding apparatus comprising a. 90. The apparatus of claim 89, wherein the decoder is Circuitry for generating class information of the coding data; And Circuitry for generating the decoded data according to the class information Data decoding apparatus comprising a. 90. The apparatus of claim 89, wherein the coding data includes prediction coefficients for generating the decoded data, and wherein the decoder includes circuitry for generating the decoded data based on the prediction coefficients and the coding data. Decoding device. 90. The apparatus of claim 89, wherein the coding data includes prediction coefficients for each class to produce the decoded data, and the decoder further comprises: Circuitry for generating class information using a plurality of samples of the coding data; And Circuitry for generating the decoded data using the prediction coefficients corresponding to the coded data Data decoding apparatus comprising a. 90. The apparatus of claim 89, wherein the decoder is A memory for storing prediction coefficients for each class; Circuitry for generating class information using a plurality of samples of the coding data; And Circuitry for reading the prediction coefficients corresponding to the class information from the memory and generating the prediction sample using the read prediction coefficients and the coding data Data decoding apparatus comprising a. 95. The apparatus of claim 93, wherein the prediction coefficients for each class stored in the memory are generated using training raw data. A method for transmitting coded image data, the method comprising: Reducing the number of pixels of the raw image data to generate compressed data, Limiting the level of pixels of the compressed data to generate limited compressed data; Correcting the limited compressed data to generate corrected data (correction step), Predicting the raw image data according to the correction data to generate prediction data having a plurality of prediction pixels (prediction step), Calculating a prediction error of the prediction data with respect to the raw image data (calculation step), Determining the suitability of the correction data as coded data based on the prediction error (judgment step), Based on the determined suitability, repeating the correction step, the prediction step, the calculation step and the determination step, and Outputting the correction data as coding data of the raw image data Receiving the coded image data generated by the; And Transmitting the coded image data Image data transmission method comprising a. 95. The method of claim 95, wherein the predicting step Generating class information of the correction data; And Generating the prediction pixels according to the class information Image data transmission method comprising a. 95. The method of claim 95, wherein the predicting step Generating prediction coefficients based on the correction data; And Generating the prediction pixels based on the correction data and the prediction coefficients Image data transmission method comprising a. 95. The method of claim 95, wherein the predicting step Generating class information based on the plurality of pixels of the correction data; Generating prediction coefficients for each class based on the raw image data and the correction data; And Generating the plurality of prediction pixels based on the prediction coefficients and the correction data corresponding to each class information Image data transmission method comprising a. 99. The method of claim 98, further comprising outputting the coding data with the prediction coefficients for each class. 95. The method of claim 95, wherein the predicting step Storing prediction coefficients for each class; Generating class information based on the plurality of pixels of the correction data; And Reading the prediction coefficients corresponding to the respective class information from the memory and generating the prediction pixels based on the read prediction coefficients and the correction data Image data transmission method comprising a. 101. The method of claim 100, wherein the stored prediction coefficients for each class are generated based on learning raw image data. 102. The method of claim 101, wherein the coding data is output together with the prediction coefficients for each class. 95. The method of claim 95, wherein the correcting step includes storing correction values, wherein the compressed data is corrected using the correction values. 97. The method of claim 95, wherein the determining of conformance Detecting whether the prediction error is less than a predetermined threshold; And Outputting the correction data as the coding data when the prediction error is smaller than the predetermined threshold value Image data transmission method comprising a.

Images